--- updatedAt: 2026-10-12T14:10:59.000Z agentTools: projectIndex: https://documentation.picsellia.com/llms.txt --- # Changelog # Changelog Picsellia SDK Python is a library that allows users to connect to Picsellia backend. All notable changes to this project will be documented in this file. ## \[6.36.1] - 2026-10-12 ### Added * Add `chunk_size` parameter to `DatasetVersion.build_coco_file_locally()` and `DatasetVersion.build_pascal_voc_locally()`, to load fewer annotations per request when loading fails or times out on big datasets. Capped and defaults to 10 for segmentation, line and multi datasets, 100 for other types. ## \[6.36.0] - 2026-10-09 ### Added * Add `Datalake.update_custom_metadata_schema()` to create, update and optionally delete keys of the custom metadata schema all at once. Unchanged keys are skipped, errors are collected per key and raised as a `CustomMetadataSchemaUpdateError` unless `raise_on_error=False`. * Add `CustomMetadataKey` pydantic models in `picsellia.types.schemas` (`StringCustomMetadataKey`, `IntegerCustomMetadataKey`, `FloatCustomMetadataKey`, `DatetimeCustomMetadataKey`, `BooleanCustomMetadataKey`). Custom metadata schema methods accept them or the corresponding dicts. * Add `Datalake.get_custom_metadata_schema()`, `Datalake.upsert_custom_metadata_schema_key()`, `Datalake.delete_custom_metadata_schema_key()` and `Datalake.set_allow_extra_custom_metadata()` to manage the custom metadata schema key by key. * Add `DatasetVersion.build_pascal_voc_locally()` to build Pascal VOC files locally and write them into a directory, avoiding platform export timeouts on huge datasets. ## \[6.35.2] - 2026-09-15 ### Added * Add `masks` parameter in `Experiment.add_evaluation()` and `MaskFormat` used in `EvaluationFormat` for `Experiment.add_evaluations()`, so you can evaluate masks. Masks are RLE encoded strings, their area and bounding box are computed by Picsellia. Evaluation metrics can then be computed with `InferenceType.MASK`. ### Fixed * Cleanup files on download errors to avoid partially downloaded files. ## \[6.35.1] - 2026-09-08 ### Fixed * File downloads are now retried on `ChunkedEncodingError` ## \[6.35.0] - 2026-09-03 ### Breaking * Removed `Processing.create_dataset_version_processing_constraints()` and `Processing.create_datalake_processing_constraints()`, deprecated since *6.29.0*. Constraints are now defined on `Processing` inputs. * `Client.create_processing()` now requires a `target` parameter, which is either `DATASET_VERSION`, `MODEL_VERSION` or `DATALAKE`. `type` is now optional. * Removed deprecated `model_version_id` and `target_datalake_name` parameters of `Datalake.launch_processing()`, and `model_version_id` and `target_version_name` parameters of `DatasetVersion.launch_processing()`. Give them into `inputs` instead. * On `launch_processing()` methods, `parameters`, `cpu`, `gpu` and `inputs` are now keyword-only arguments. ### Added * Add `ProcessingTargetType` enum and `Processing.target` property * Add `sizes` in `Datalake.upload_data()`, so you can give the width and height of files we cannot open ourselves, such as hdf5 files ## \[6.34.0] - 2026-07-16 ### Added * Add `Mask` to handle semantic datasets. * Add `Annotation.create_mask()`, `create_multiple_masks()` and `list_masks()` to create or list masks * Add `DatalakeChart` it allows you to manipulate charts with the sdk * Add `Datalake.create_chart()` and `Datalake.list_charts()` to create and list charts * Add `ComputeConnector`, you can now choose where to launch experiments ### Changed * `list_data()`, `list_assets()`, `list_predicted_assets()` and `list_evaluations()` will list items with a cursor in some cases, which will improve performance for large datalakes, datasets, deployment or experiments. ### Fix * There were too much retries on `monitor()` and `predict()` methods ## \[6.33.0] - 2026-06-09 ### Added * Add Keypoint shape type for deployment monitoring * Add Mask shape type for deployment monitoring * Add `Client.get_public_processing()` to retrieve public processing * Add `export_key` in `DatasetVersion.export_annotation_file()`, you can now export annotation files with different key values (object\_name, filename, asset id or data id) * Add `export_key` in `DatasetVersion.build_coco_file_locally()`, exported coco file will have different key values * Add `export_key` in `DatasetVersion.import_annotations_coco_file()`, so you can use exported annotation file to import your annotations with different key values * Add `Datalake.update_multiple_data()`, you can update multiple data with this method * Add `custom_metrics` in `Experiment.add_evaluation()`, so evaluation will hold custom metrics that you can filter on or create charts with * Add `Experiment.add_evaluations()`, to create evaluations on multiple assets at the same time * On `Experiment.launch()` you can specify `gpu_type` to use a different GPU than default for the training * Add `object_name`, `filename`, `url` properties on Assignment classes * Add `AnnotationCampaign.get_campaign_issue_counts()` and `ReviewCampaign.get_campaign_issue_counts()` to return issue counts for a list of users ### Changed * On `requests.ConnectionError`, api calls to monitoring services will retry up to *5* times with an exponential backoff. It means it will wait up to 7.5 seconds. ## \[6.32.0] - 2026-04-29 ### Breaking * `Deployment.add_review()` does not allow string anymore, but `Label` object ### Added * InferenceType `MASK` is now allowed for `DatasetVersion` * `DatasetVersion.import_coco_file()` will accept RLE masks in `DatasetVersion` of type MASK. At the moment, we only allow already encoded strings * `Deployment.monitor()` allows `custom_metrics` parameter, that will be stored on `PredictedAsset`. * Add `DatasetVersion.similar_search()` and `DatasetVersion.text_search()` to query assets based on our embeddings search service * Add `Assignment` object * Add `Campaign.list_assignments()` to list `Assignment` objects * Add `Campaign.reassign()` to reassign `Assignment` to users * Coco file might have categories with `keypoints` as a string, we will accept it and create a skeleton with only one point ### Changed * We deprecated `Datalake.visual_search()` and renamed it to `Datalake.similar_search()` * All requests will have now a default timeout of 300s * Lowering chunk size on load annotations for dataset SEGMENTATION and MULTI, to handle really large polygons (this method is used on build\_coco\_file\_locally) ### Fixed * Exif metadata was not handling well null bytes ## \[6.31.0] - 2026-03-31 ### Added * `inputs` parameter must be given to launch new processing * Add `attributes` property on `Annotation` and shapes * Export coco file and build\_coco\_file() will return `attributes` key filled with attributes of shapes and assets * Import coco file with attributes will be sent to platform and saved * `get_attributes()` and `set_attributes()` can be called on a DatasetVersion or a Label to handle attributes configuration * `postprocess` boolean can be set on all `predict()` methods to receive a fast but un-postprocessed response from our serving service * `launch_dataset_processing()` now allows `assets` to have access to a list of asset id in your processing * `Deployment.list_labels()` can be called to retrieve labels of a deployment. You won't be able to update their name or delete them, because labelmap is immutable. ### Changed * We will retry get and xget methods on a ChunkedEncodingError, that could happen on long list\_annotations and build\_coco\_file\_locally * Renaming `Connexion` to `Connection` class, also renamed files. ### Breaking * `Label` are now instantiated either with a deployment\_id or a dataset\_version\_id. You were not supposed to create an instance by hand, if you did, make sure you add `dataset_version_id` as a parameter of the constructor ### Build * We now use `uv` to build the package, that should not change anything for you ## \[6.30.2] - 2026-03-20 ### Fixed * `DatasetVersion.import_annotations()` could failed for keypoints * `AnnotationCampaign.update()` parameter `annotated_entry_step` now allows `DONE`. On `sync_assets()`, asset with annotation will be sent directly to DONE. ## \[6.30.1] - 2026-03-03 ### Fixed * `Model.create_version()` did not allow `KEYPOINT` type. ## \[6.30.0] - 2026-02-13 ### Added * `DatasetVersion.build_coco_file_locally()` now supports keypoints (skeletons) export. ## \[6.29.0] - 2026-01-22 ### Added * `Deployment` and `ModelVersion` with type `LINE` and `POINT` are now allowed. * `LinePredictionFormat` and `PointPredictionFormat` can be used to monitor * `PredictedAsset.add_review()` is possible with `points` and `lines` * `JobRunStatus.QUEUED` has been added, it is used by Myoboku to handle job picked but not running yet * `Deployment.monitor()` has a new parameter called `sample`. If it's given, you can choose whether to send or not the prediction to the platform. If it's not set, the system will rely on sample\_rate if defined, else it will be sampled. * `ReviewCampaign.close()` and `AnnotationCampaign.close()` can be called to close a campaign * `ReviewCampaign.reopen()` can be called to reopen a campaign * `Project.create_experiment()` can be called with parameter `parameters` to create the Log parameters. ### Fixed * `Project.create_experiment()` with `base_experiment_id` and `base_model_version_id` were not attaching logs and artifacts. ### Changed * `Datalake.import_bucket_objects`, `Datalake.import_cloud_objects` and `Datalake.import_cloud_projections` are calling a new api on datalake object * `Processing.create_processing_constraints` allows str and None ### Deprecated * `Experiment.update()` with parameter `base_experiment_id` and `base_model_version_id` were not doing what they were supposed to. So those parameters are deprecated. * `Experiment.attach_model_version()` is now deprecated. * You can call `Experiment.set_base_experiment()` and `Experiment.set_base_model_version()` instead ### Breaking * We removed dependency of `picsellia-connexion-services`, and renamed classes from Connexion to Connection. We will rename all connexion to connection in the next release, making it retro-compatible. * `Point.order` is not used in the platform anymore. Some methods that were expecting order have been broken ## \[6.28.1] - 2025-12-12 ### Changed * `DataProjection.type` is deprecated, projections are only user projections * `Data.list_projections()` is not using filter type anymore ### Fixed * `split_into_multi_assets()` was not returning label distribution when keypoints were used ## \[6.28.0] - 2025-12-10 We added a new InferenceType (available across the whole platform) for skeletons/keypoints handling. This InferenceType is named KEYPOINT (not to confuse with POINT that is used to draw single points where each point has a Label). As of today, we only support storing KEYPOINT annotations, exporting them in COCO files and storing evaluations. This is still EXPERIMENTAL and no visualization features has been developed yet so you can only manipulate those KEYPOINTS with the SDK for now. ### Added * `InferenceType.KEYPOINT` is now available for DatasetVersion and ModelVersion * `ModelVersion.update()` do not block parameter `type` for LINE, KEYPOINT or POINT * `DatasetVersion.import_annotations_coco_file()` will handle keypoint files if DatasetVersion is of type KEYPOINT * `DatasetVersion.export_annotation_file("COCO")` will handle keypoint files if DatasetVersion is of type KEYPOINT * `Label.get_skeleton()` can get skeleton of your Label * `Label.set_skeleton()` can set up a skeleton on your Label * `Experiment.add_evaluations()` allows parameter `keypoints` to create evaluations ### Changed * `picsellia-connexion-services` package has been updated with a new minimal version of `requests` package, because of a security flaw ## \[6.27.1] - 2025-12-03 ### Added * `Experiment.get_attached_dataset_versions()` can be called to fetch attached dataset version as a dict with aliases ## \[6.27.0] - 2025-12-01 Happy Advent of Code y'all Instead of gifts, we dropped python3.9 for Christmas, and allowed python3.14 in this patch ### Added * `DatasetVersion.get_asset_tag_analytics()` can be called to fetch usage of asset tags in a DatasetVersion * `DatasetVersion.get_data_tag_analytics()` can be called to fetch usage of data tags in a DatasetVersion * `DatasetVersion.lock()` can be called to lock a dataset * `DatasetVersion.unlock()` can be called to unlock a dataset * Allow python3.14 * While dropping python3.9, we could get rid of old syntaxes like type Union and Optional ### Changed * `reset_url()` internal methods of Data, Asset, PredictedAsset and DataProjection will call a more optimized endpoint, but won't refresh object anymore ## \[6.26.2] - 2025-10-30 ### Added * `Deployment.list_predicted_assets()` now allows `q`, `assignment_status`, `assignment_step_id`, `assignment_user_id` and `custom_metadata` * `DatasetVersion.list_assets()` now allows `custom_metadata` * `PredictedAsset.get_prediction()` can be called to retrieve data of its prediction * `PredictedAsset.get_shadow_prediction()` can be called to retrieve data of its shadow prediction * `PredictedAsset.get_review()` can be called to retrieve the last review created on this `PredictedAsset` ## \[6.26.1] - 2025-10-21 ### Added * `Datalake.import_cloud_objects()` can be called with metadata ### Changed * `Deployment.predict()` and `monitor()` methods will retry on errors with status code greater or equal than 501, or in case of common connection errors ## \[6.26.0] - 2025-10-21 ### Added * `Datalake.import_cloud_objects()` can be called to import objects from your storage into your datalake with custom metadata, source and tags on each data. * `Datalake.import_cloud_projections()` can be called to import (DataProjection) linked to existing data of your Datalake. * `Deployment.predict()` and `monitor()` now allows `custom_metadata` ### Changed * `DatasetVersion.build_coco_file_locally()` will always build a coco file with all categories. ## \[6.25.0] - 2025-09-26 ### Added * `Datalake.upload_data()` now allows a `custom_metadata` parameter, this can be an arbitrary dict that will be attached to your Data. * `Datalake.list_data()` allow `custom_metadata` parameter. You can filter your data based on the custom\_metadata field. * `Data` now has a `custom_metadata` property. * `Data.update_custom_metadata()` can be called to update the custom\_metadata field of a Data * `Experiment` now has a `project_id` property ## \[6.24.1] - 2025-09-16 ### Fixed * `_create_projection_from_cloud_image()` was not allowing empty metadata ## \[6.24.0] - 2025-07-28 ### Added * `DatasetVersion.list_assets()` can be called with parameters `assignment_status`, `assignment_user_id` and `assignment_step_id`, to filter on assets in campaign. * `DatasetVersion.list_import_coco_file()` will create LabelGroup if field `supercategory` of categories object are set, and that force\_create\_label is True #### DataProjection * DataProjection objects represent projections of Data on the platform. They can be downloaded or deleted. * `Datalake.create_projection()` now returns a DataProjection * `Data.list_projections()` can be called to retrieve DataProjection objects linked to a Data #### LabelGroup * LabelGroup objects represent groups of labels on the platform. * `LabelGroup.get_parent()` can be called to retrieve the parent of the group * `LabelGroup.set_parent()` can be called to set the parent of the group * `DatasetVersion.create_label_group()` can be called to create a label group * `DatasetVersion.list_label_groups()` can be called to list all label groups of a dataset * `Label.get_group()` can be called to retrieve the group of a label * `Label.set_group()` can be called to set the group of a label ### Changed #### Worker Deprecation As we are moving to a new user management system, the Worker class will progressively be deprecated. The Worker class existed before our Annotation Campaign system to create multiple annotations on assets. Since it is no longer possible to create multiple annotations without the campaign system, we can remove many filters and parameters from SDK methods. **Methods that will no longer use the given worker parameter:** * `Annotation.update()` * `Asset.get_annotation()` * `Asset.create_annotation()` * `DatasetVersion.build_coco_file_locally()` * `DatasetVersion.export_annotation_file()` * `DatasetVersion.import_annotation_*_file()` * `DatasetVersion.list_annotations()` * `DatasetVersion.load_annotations()` * `Experiment.compute_evaluation_metrics()` **Methods that will raise an error if a worker is provided:** * `Asset.delete_annotations()` * `MultiAsset.delete_annotations()` * `DatasetVersion.delete_all_annotations()` **Important note:** `Worker.id` will no longer be the real worker ID. It will be either a UUID representing 0 or a user ID. You should not use it anymore. **Migration to User class:** Do not use the Worker class anymore. Instead, use the User class, which manipulates real users of the platform. * `Dataset.list_workers()`, `Dataset.find_worker()`, and `Deployment.list_workers()` will return Worker objects but should not be used anymore * Instead, use `Dataset.list_users()` and `Deployment.list_users()` which have been added * You can also use `Client.list_users()` to get users from your organization * You can call `Dataset.add_user()` and `Deployment.add_user()` to add a User to a resource. This is useful for UNPRIVILEGED users that you want to add as LABELER on campaigns * In `Campaign.add_step()` and `Step.update()`, `assignees` parameters now accept a User and a user ID ## \[6.23.0] - 2025-07-01 ### Added * `Datalake.upload_data()` will accept parameter `upload_dir` when using a private object storage. You can now specify the directory path on the object storage * `Deployment.monitor()`, `Deployment.monitor_bytes()`, `Deployment.predict()` and `Deployment.predict_bytes()` will also accept this `upload_dir` parameter * `Processing.create_dataset_version_processing_constraints()` can be called to create constraints on dataset version processings * `Processing.create_datalake_processing_constraints()` can be called to create constraints on datalake processings ### Fixed * `Data.load_embeddings` and `Asset.load_embeddings` were not working anymore due to a minor change in api * When reading exif data of images, resolution\_unit was not correctly read. ## \[6.22.0] - 2025-06-16 ### Added * `DatasetVersion.start_fast_training()` to start a fast training on a DatasetVersion * `DatasetVersion.list_fast_trainings()` to get `FastTraining` objects linked to this DatasetVersion * `FastTraining.list_evaluations()` to get evaluations done during this training * `FastTraining.list_logs()` to list logs done during this training * `FastTraining.get_log()` to get a log done during this training * `FastTraining.get_parameters()` to get a parameters used by this training * `FastTraining.list_artifacts()` to get files of the model version used by this training * `Deployment.update()` allows parameter `sample_rate` which can be used to not send all images to platform, but still computing monitoring data * `DatasetVersion.launch_processing()` now allows to select `model_version_id` and `target_version_name` used during processing * `Datalake.launch_processing()`now allows to select `model_version_id` and `target_datalake_name` used during processing ## \[6.21.8] - 2025-05-16 ### Changed * On `requests.ConnectionError`, all api calls will retry up to *5* times with an exponential backoff. It means it will wait up to 7.5 seconds. ## \[6.21.7] - 2025-05-14 ### Added * AssetLister and DataLister now have a `list_ids()` method. This is an internal method that should be used with care and can change in the future ## \[6.21.6] - 2025-05-14 ### Changed * On `requests.ConnectionError`, all api calls will retry up to 4 times with an exponential backoff ## \[6.21.5] - 2025-04-17 ### Added * `DatasetVersion.compute_shapes_embeddings()` will wait for job to end by default * `DatasetVersion.activate_visual_search()` will wait for job to end by default ## \[6.21.4] - 2025-04-15 ### Experimental * Features for shape embeddings on DatasetVersion were added ### Added * `Datalake.embeddings_computation_status()` to fetch status of embeddings computation on Datalake * `DatasetVersion.embeddings_computation_status()` to fetch status of embeddings computation on DatasetVersion * `DatasetVersion.compute_shapes_embeddings()` to start computation of shape embeddings on DatasetVersion * `DatasetVersion.delete_shape_embeddings()` to delete computation of shape embeddings on DatasetVersion * `DatasetVersion.get_shapes_embeddings_status()` to fetch status of shape embeddings computation on DatasetVersion * `DatasetVersion.list_rectangles_embeddings()` to fetch rectangle embeddings on DatasetVersion * `DatasetVersion.list_polygons_embeddings()` to fetch polygon embeddings on DatasetVersion ## \[6.21.3] - 2025-04-09 ### Added * To prepare next release of picsellia-agents, we need some internal methods ## \[6.21.2] - 2025-03-31 ### Changed * Upload of large files will be retried 3 times on Read Timeout * Better handling of large annotation files on all `import_annotations_file` methods. ## \[6.21.1] - 2025-02-24 ### Fixed * An annotation type hint was causing bug with python >= 3.11 ### Docs * Some minor doc fixes ## \[6.21.0] - 2025-02-20 Some features around Visual Search ! ### Changed * Remove `detection_boxes` in SegmentationPredictionFormat ### Added * `Datalake.visual_search()` can be called to search similar data * `Datalake.text_search()` can be called to search data that match a text query * `Datalake.count_embeddings()` can be called to count number of data indexed * `Datalake.list_embeddings()` can be called to list embeddings of in your datalake * `DatasetVersion.activate_visual_search()` can be called to activate visual search in your dataset * `DatasetVersion.deactivate_visual_search()` can be called to deactivate visual search in your dataset * `DatasetVersion.count_embeddings()` can be called to count number of data indexed in your dataset * `DatasetVersion.list_embeddings()` can be called to list embeddings of in your dataset * `Data.embeddings` property can be used to retrieve stored embeddings * `Asset.embeddings` property can be used to retrieve stored embeddings ### Fixed * `Deployment` attribute `_review_campaign_id` was not well-defined in some cases ## \[6.20.0] - 2025-01-20 Happy new year you all ! Some new features around campaigns, some refactoring, some breaking changes in this version as deprecated code was removed. ### Breaking Behavior has changed for users that initialize Client() without giving `organization_name` or `organization_id` * Users that belong to only one organization will be connected to this organization. * Users that belong to multiple organizations but own one will be connected to owned organization. * Users that belong to multiple organizations but do not own any organization won't be able to initialize an unspecified Client. * `AnnotationCampaign.list_steps()` will return an AnnotationCampaignStep instead of a dict! * `Model.type` and `Model.framework` were deprecated and are removed from now. * `Datalake.find_all_datas()` was deprecated and is removed from now. You should use `list_data()` instead. * `DatasetVersion.find_all_assets()` was deprecated and is removed from now. You should use `list_assets()` instead. * `ModelContext.retrieve_experiment()` was deprecated and is removed from now. You should use `get_experiment()` instead. * `Deployment.check_feedback_loop_status()` was deprecated and is removed from now. * `Deployment.disable_feedback_loop()` was deprecated and is removed from now. * `Deployment.setup_continuous_training()` will not use deprecated parameter `dataset_version` anymore. * `Experiment.publish()` was deprecated and is removed from now. You should use `export_as_model_version()` instead * `ModelVersion.update_thumbnail` was deprecated and is removed from now. You should use the web app instead. ### Added * `AnnotationCampaignStep` model, to manage your campaign step * `Client.get_processing()` to find a processing by its name * `AnnotationCampaign.sync_assets()` can be called to add assets that are not assigned as task in your campaign * `AnnotationCampaign.update()` can be called with `default_entry_step` and `annotated_entry_step` parameter * `AnnotationCampaignStep.redistribute_assignments()` can be called to add assets that are not assigned as task in your campaign * `AnnotationCampaignStep.update()` can be called to update a step * `AnnotationCampaignStep.delete()` can be called to add assets that are not assigned as task in your campaign * `Client.list_annotation_campaigns()` can be called to list annotation campaigns of your organization * `Client.list_review_campaigns()` can be called to list review campaigns of your organization * `Worker` class has now a type : `DATASET` or `DEPLOYMENT` * `Worker.user_id` property * `Deployment.list_workers()` can be used to list workers of a deployment * `ReviewCampaign` and `ReviewCampaignStep` models can be used to handle campaigns on Deployment ! Methods are more or less the same, except for some edge case behaviors. * `Client.create_processing()` can be called with `description` parameter * `DatasetVersion.copy_assets_to()` can be called to copy assets from a dataset to another ### Changed * `DatasetVersion.list_assets()` should behave better when a list of parameter is given * `Datalake.list_data()` should behave better when a list of parameter is given * `DatasetVersion.add_data()` will add data by batch, so it won't fail with 413 error when more than 300k assets are given * `DatasetVersion.add_data()` will now wait for job to end by default * `DatasetVersion.fork()` will now wait for job to end by default * `AnnotationCampaign.create_step()` can be called with `sample_rate` parameter ### Removed * `AnnotationCampaign` do not require `name` anymore ### Fixed * `ModelVersion` str method was not displaying well its origin name * Handle 413 errors (when payload or uri are too large) * Some bad generated documentation should be fine now * Some methods were not reading filenames correctly on Windows ## \[6.19.2] - 2024-11-06 ### Added * `DatasetVersion.find_asset()` can be called with `id` parameter to find an asset by its id. ## \[6.19.1] - 2024-10-31 Happy Halloween ! ### Added * `Deployment.find_predicted_asset()` to find a PredictedAsset from different criteria ## \[6.19.0] - 2024-10-30 We dropped support of python3.8, goodbye old friend, you will be missed. ### Added * `Deployment.predict_shadow()` to predict with shadow model on an already processed PredictedAsset * `Deployment.monitor_shadow()` and `Deployment.monitor_shadow_from_oracle_prediction_id()` to monitor with shadow model on an already processed PredictedAsset * Added support of python3.13 * Bumped some packages * While dropping python3.8, we could get rid of old syntaxes * Better handling of conflict when monitoring already processed data * A GitHub workflow so we can send to our Slack channel a release note ### Changed * We welcomed ruff as a new pre commit linter * As we do not allow multi campaign on the same dataset, we changed `DatasetVersion.get_campaign()` ### Fixed * We don't warn anymore if you have a future version of the SDK ### Deprecated * We dropped support of python3.8 ## \[6.18.3] - 2024-10-01 ### Added * New processing types allowed on the platform : data auto tagging, model compression, model conversion ### Fixed * Deployment.predict() and Deployment.predict\_bytes() were failing when sending metadata ## \[6.18.2] - 2024-09-17 ### Fixed * We refactored upload of large file to allow upload very large file on Google cloud storage ## \[6.18.1] - 2024-09-05 ### Fixed * Added `Asset.content_type`, needed in `Asset.to_data_schema()` ## \[6.18.0] - 2024-09-04 ### Added * `Deployment.predict_cloud_image()` can be called to predict onto an image stored on your object storage * `Deployment.predict_data()` can be called to predict onto a data stored on your target datalake * `Deployment.monitor_cloud_image()` can be called to monitor an image stored on your target datalake * `Deployment.monitor_data()` can be called to monitor a data stored on your target datalake * All `Deployment.predict()` can be called with `metadata` dict parameter. It will be stored on Data after being processed. * `Deployment.monitor()` can be called with `metadata` dict parameter aswell. * All `Deployment.predict()` can be called with `monitor` boolean parameter. It allows you to only call inference service, it won't call our monitoring service. ## \[6.17.1] - 2024-08-22 ### Fixed * On `DatasetVersion.build_coco_file_locally()` area of segmentation will now be computed properly ## \[6.17.0] - 2024-07-29 ### Added * `Experiment.launch()` now return a Job * `PredictedAsset` class can be used to manage predictions of a Deployment. * `PredictedAsset.add_review` class can be used to add review to assets of a Deployment. * `Deployment.list_predicted_assets()` can be used to list assets of a Deployment. * `Processing` class can be used to manage your processings * `Client.list_processings()` can be used to list your processings. * `DatasetVersion.launch_processing()` can be used to launch a processing on your DatasetVersion. ### Fixed * On `Datalake.upload_data()` with parameter fill\_metadata=True could not pass validation of our api in some case, so we need to cast values. * `DatasetVersion.list_assets()` with parameter filename\_startswith was not working * `DatasetVersion.build_coco_file_locally()` will only use the LAST annotation created on each asset. In campaign mode it caused some problems ### Changed * Some apis were deprecated on the backend side so we changed it on the SDK. These apis will be removed in the future : SDK versions prior to 6.17.0 might break. * Allowed version of Pillow is now ">=9.2.0, <11" ## \[6.16.0] - 2024-06-05 ### Added * `DatasetVersion.list_assets()` can now be called with a list of data ids * `DatasetVersion.import_annotations_coco_file()` can now be called with `use_id`. If you exported your coco file with `use_id`, you can use import it now. * `DatasetVersion.import_annotations_coco_video_file()` can be called to import video annotations in your dataset. You could export it from picsellia with the SDK since the 6.13.0 * `Rectangle`, `Classification`, `Polygon`, `Line`, `Point` have now `text` field that you can use as an OCR feature. You can call `update` on shape to update text field. * `DatasetVersion.import_annotations_coco_file` and its video counterpart will read `utf8_string` field of json coco file and import it into the platform. ### Fixed * On `Datalake.upload_data()` with parameter fill\_metadata=True could not pass validation of our api in some case, so we need to cast values. * `DatasetVersion.list_assets()` with parameter filename\_startswith was not working * `DatasetVersion.build_coco_file_locally()` will only use the LAST annotation created on each asset. In campaign mode it caused some problems ### Changed * We upgraded pydantic dependency : we now depends on its version 2 * Default max pagination is reduced to 1000. It was 10000, it's a little bit too much * We reduced to 1000 the max chunk\_size on `load_annotations()` * We refactored tests and their files ### Deprecated * We dropped support of python3.7 * All stuff related to Scan have been removed ## \[6.15.0] - 2024-05-06 ### Added * `Datalake.import_bucket_objects()` can now be called with `tags` and `source`, that will be given to created data. ### Fixed * Typo on get campaign log * `wait_for_upload()` had 2 beartype decorators * Some tests were failing ## \[6.14.2] - 2024-04-19 ### Changed * We now use session object to download and upload from presigned url. This should speed up a little bit, and not create enormous amount of connection. ## \[6.14.1] - 2024-04-16 ### Added * `Datalake.list_data()` can be called with ids to fetch data from their ids. ### Fixed * When listing data and assets, filtering with tags and filenames was not working * Downloadable objects such as Data, Asset and Artifact, were syncing before downloading, we don't want to do that anymore * Type hint problem on AnnotationCampaign * Documentation on AnnotationCampaign * Tests on AnnotationCampaign ## \[6.14.0] - 2024-03-19 ### Added * `DatasetVersion.create_campaign()` can be used to create an AnnotationCampaign. * `AnnotationCampaign.add_step()` can be used to add a step to an AnnotationCampaign. * `AnnotationCampaign.launch()` can be used to add assignments to all assets on this campaign of your DatasetVersion. * `AnnotationCampaign.update()` can be used to update an AnnotationCampaign. * `DatasetVersion.get_campaign()` can be used to retrieve the AnnotationCampaign of this DatasetVersion. * `DatasetVersion.export_annotation_file()` can be called with parameter `use_id=True`. Generated files will use asset id insted of filename. * `Datalake.import_bucket_objects()` can be used to import bucket objects of your S3 that are not yet on the platform. ### Fixed * Calling `MultiAsset.as_multidata()` with more than 10k assets was bugged * Some dependencies were fixed on pyproject.toml, such as pyyaml, beartype and orjson. Python 3.11 and Python 3.12 can now be used with the sdk ## \[6.13.0] - 2024-02-26 ### Added * `Datalake.upload_data()` can now handle unreadable images (like hdf5) and video ! By default, it will wait for data to be full processed by our services * `Datalake.create_projection(data)` can be used to create a projection on a data : you have to give a name for this projection, it will be of type CUSTOM. * `DatasetVersion.export_annotation_file()` can be called with boolean parameter `export_video` to only export video in annotation file ### Fixed * Default content type assigned to data we cannot read mime type is now application/octet-stream ## \[6.12.0] - 2024-01-17 ### Added * `Data`, `MultiData`, `Asset`, `MultiAsset` can be downloaded with `download(use_id=True)` to use id as filename. Extension of base filename will be used. * `DatasetVersion.build_coco_file_locally()` can now be called with parameter `use_id=True` to build a coco file with `file_name` keys as `.` * When initializing a Client, you can pass parameter `session` with your own requests.Session object, that will be used on each request done by the sdk. It allows configuring a proxy or custom headers. ## \[6.11.0] - 2024-01-10 Happy new year everyone ! Some minor features and fixes in this version ### Added * `Data.update_metadata()` can be called to update metadata of a data. * `DatasetVersion.load_annotations()` can now be called with parameter `assets` to build a json coco file with only given assets. ### Fixed * When downloading a file, there was a race condition when creating parent directories of downloaded file. * Storing an artifact with `do_zip=True` was not creating expected archive name when a dot was present in filename. * On upload, large file discriminant is now 25MiB and not 5MiB ### Changed * Major changes in docstrings, a lot of fixes! ## \[6.10.3] - 2023-09-22 ### Fixed * PredictionFormat will not raise an exception if one of its list is empty. `Deployment.monitor()` can now be called with empty predictions. ## \[6.10.2] - 2023-09-06 ### Fixed * `Deployment.predict_bytes()` needs parameter `filename` to work. It was sending prediction with filename `media`. ## \[6.10.1] - 2023-08-31 ### Added * `DatasetVersion.build_coco_file_locally()` now adds area of bounding boxes on generated files for Object Detection. ## \[6.10.0] - 2023-08-21 Some deprecation on privacy object creation, it is not possible to create a public dataset, project or model from the sdk anymore. Attach dataset version to Feedback Loop and Continuous Training settings of your deployments. ### Added * `Deployment.attach_dataset_to_feedback_loop()` can now be called to attach a dataset to Feedback Loop settings of your deployment. Attached datasets can be filled with reviewed prediction from your deployment. * `Deployment.detach_dataset_from_feedback_loop()` allows detaching dataset from Feedback Loop. * `Deployment.list_feedback_loop_datasets()` will list attached datasets of your Feedback Loop. * `Deployment.attach_dataset_to_continuous_training()` can now be called to attach a dataset to Continuous Training settings of your deployment. Attached datasets will be added to your experiment created when Continuous Training is triggered. * `Deployment.detach_dataset_from_continuous_training()` allows detaching dataset from Continuous Training. ### Deprecated * Parameter `dataset_version` of `Deployment.setup_feedback_loop()` * Parameter `dataset_version` and `model_version` of `Deployment.setup_continuous_training()` * Parameter `private` of `Client.create_model()` * Parameter `private` of `Model.update()` * Parameter `private` of `Client.create_project()` * Parameter `private` of `Project.update()` * Parameter `private` of `Client.create_dataset()` * Parameter `private` of `Dataset.update()` ## \[6.9.0] - 2023-07-10 ### Added * `Datalake.upload()` can now be called with parameter `fill_metadata`. On upload, set this parameter to True to read exif metadata flags and push metadata to Picsellia. By default, `fill_metadata` is False. * `Client.get_datalake()` can now be called with parameter `id` or `name` to retrieve the Datalake you want. If nothing is given, this method will retrieve your default datalake. * `Deployment.predict_bytes()` can now be called if you want to send image as bytes on Serving service * `ModelVersion.update()` and `Model.create_version()` can now be called with `docker_tag` parameter ## \[6.8.0] - 2023-06-12 ### Added * `Datalake.upload_data()` now accept parameter `metadata` which is a dict (or a list of dict if there are multiple filepaths) matching common metadata allowed by Picsellia. * `DatasetVersion.fork()` can now be called with parameters `with_annotations` and `with_labels` as web application * `DatasetVersion.export_annotation_file()` can now be called with `AnnotationFileType.YOLO` to retrieve a zip with yolo files * `Dataset.create_model()` and `DatasetVersion.update()` have parameter `description` to update description of a ModelVersion ### Fixed * When adding evaluation, allow empty list to be sent as a list of shape ## \[6.7.1] - 2023-05-15 Some changes on `monitor()` following an update of our monitoring stack. ### Added * Property `type` of a deployment. * Method `Deployment.monitor_bytes()` can be called if you have an image as bytes to send to monitoring. ### Changed * `Deployment.monitor()` do not use parameters `model_version` and `shadow_model_version` anymore * Parameter `shadow_raw_predictions` of `Deployment.monitor()` will be replaced by parameter `shadow_prediction` in the future * You can add `content_type` parameter to monitor() if you don't want sdk to infer it with mimetypes. It will be checked with common content types of the mimetypes library * `content_type` of monitor\_bytes() is mandatory and should be a SupportedContentType enum or string. Current common type supported are "image/png" and "image/jpeg" ## \[6.7.0] - 2023-05-05 ### Added * Add `Client.create_deployment()` to create a deployment. Allow user to create it without using Picsellia Serving * Add `DatasetVersion.retrieve_stats()` to retrieve stats of your dataset version * Add `DatasetVersion.train_test_val_split()` to split a dataset version into 3 different multi assets * Add `DatasetVersion.split_into_multi_assets()` to split a dataset version into N multi assets and return their label repartition * Add `MultiData.split()` and `MultiAsset.split()` to split from a given ratio a multi asset * Add User-Agent with picsellia version in headers of requests ### Changed * Framework and type of Model are now configurable into ModelVersion * Get by id now use parent API to ensures object are in the same organization as the one connected * Methods manipulating tags on client are now calling other routes * Some minor fixes on documentation ### Fixed * Segmentation format used in `monitor()` was not supported by monitoring service ## \[6.6.0] - 2023-04-06 ### Added * `list_data` and `list_assets` have a new parameter `q` that can be used the same way the query language is used in the web platform * Deployment has new methods: `set_training_data`, `check_training_data_metrics_status` and `disable_training_data_reference`, that can be used for monitoring and unsupervised metrics. * `as_multidata` of MultiAsset can now be called with parameter * Artifact, Data, Asset, LoggingFile, ScanFile, ModelFile are now inheriting from Downloadable, and have `url` property that can be used to download files. These urls are presigned and expired at some point in the future. * Methods `add_evaluation`, `list_evaluations` and `compute_evaluations_metrics` of Experiment can be used to add, list and compute evaluation of an Experiment ### Changed * Deployment Feedback Loop dataset is now only used as a recipient for new assets submitted after review in the predictions dashboard * bbox of COCO annotation cannot be a tuple anymore ## \[6.5.0] - 2023-03-15 ### Added * Jobs are now handled differently in Picsellia platform * `get_dataset_version_by_id` can be done in Client * `get_model_version_by_id` can be done in Client * `get_experiment_by_id` can be done in Client * Import yolo files with `import_annotations_yolo_files` * `upload_data` can be called with an ErrorManager to retrieve errors after upload. ### Fixed * Pascal VOC files parsing allows some fields to not be set * 502 errors were not handled correctly * Uploading images that were transposed now correctly send width and height on the platform ### Changed * Downloading files has a new retry policy * When importing YOLO, COCO or Pascal VOC files, you need to set type of dataset before. * Some refactor on import annotations, it should be faster now ! ## \[6.4.2] - 2023-03-03 ### Changed * Return line after logging chapter * Allow parameter `enforced_ordered_categories` on `build_coco_file_locally()` to enforce categories of built coco file. ### Fixed * Do a paginated call in dataset version `train_test_split` when loading assets * Method `set_type` of dataset version was failing when logging result if a string was given ## \[6.4.1] - 2023-02-17 ### Fixed * Some methods were using types not compatible with python older than 3.8 ## \[6.4.0] - 2023-02-16 ### Added * DatasetVersion `list_assets()` will call XGET method when filtering on filenames or object\_names, to not have the limit size query error. ### Fixed * Datalake `upload_data()` now allow effectively a Datasource * Import annotations was limited to 10 000 annotations because `find_assets()` was used. It is now using `list_assets()` which is paginated. ### Changed * Deployment `update()` cannot change active anymore. ### Deprecated * DatasetVersion `find_all_assets()` ### Experimental * DatasetVersion `build_coco_file_locally()` will create a COCOFile object that can be written in a json file ## \[6.3.2] - 2023-01-31 ### Added * `Experiment.export_in_existing_model()` method will create a version of given model when exporting experiment. * Datalake `list_data()` can be used to filter on object\_names. ### Fixed * `ModelVersion.store()` was not using `replace` parameter correctly, it will now delete existing model file if it exists ### Changed * All `max_workers` parameters are now `None` by default, to use cpu count + 4 as default value * Datalake `list_data()` will call XGET method when filtering on filenames or object\_names, to not have the limit size query error. ### Removed * `AnnotationStatus.REVIEWED` status was never really used and is now removed from Picsellia. ## \[6.3.1] - 2023-01-20 This is a patch to fix download dataset on huge datasets ### Fixed * `as_multidata` now uses xget to retrieve all datas with given list of ids ### Changed * `as_multidata` is not called in `download` from DatasetVersion ## \[6.3.0] - 2023-01-19 Happy new year from the Picsellia Team ! This minor version add some useful methods and fix specified COCOFile format. ### Added * List public models with `list_public_models()` from the Picsellia Hub in Client. * Find public model with `find_public_model()` from the Picsellia Hub in Client. * Convert MultiAsset into MultiData with `as_multidata()` and `as_list_of_data()` * Allow asset tags to be added when adding data to a dataset with `add_data()` * Method `__repr__` has been added to MultiObjects * Property `labels` in ModelVersion objects * Add `get_dataset` in ModelContext to get DatasetVersion from its attached name ### Fixed * **breaking**: COCOFile was wrongly specified: in annotations, only one segmentation was allowed for each shape, but COCOFile allows multiple polygons. Now you need to give a list of list of coordinates instead of only one list of coordinates. `import_annotations_coco_file()` should now work fine with COCO files with segmentation. * Add AuthenticationError again in exceptions, which was used by picsellia-tf2 ### Changed * **may be breaking**: Default value of force\_create\_label is now True in `import_annotations_coco_file` and `import_annotation_voc_file`. * Rename `retrieve_experiment` into `get_experiment` in ModelContext * `retrieve_datasets` in ModelContext now return a dict with attached name as key and dataset version as value * `download()` a DatasetVersion is not calling assets extended endpoints which was taking a lot of time on huge datasets. * Imports order with pre-commit cln tool * Doc generation with all properties docstring ## \[6.2.1] - 2022-27-12 ### Fixed * Fixed .predict() behavior by removing mandatory parameters tags and source ## \[6.2.0] - 2022-27-12 ### Added * Add docker env variables support on ModelVersion * Add an experiment.log\_parameters() to use experiment.log() with parameters more easily * Add possibility to attach base parameters when experiment.attach\_model() * Add feedbackloop method check status * Add `__repr__` methods to DAO objects, for better UX representation * Add possibility to set tags and source when sending a prediction to serving * Add get\_or\_create\_datasource() in client ### Changed * Rename experiment.publish() into experiment.export\_as\_model() * Some changes on feedbackloop methods * Method list\_data and list\_assets can now use parameter `intersect_tags` to find objects that have all tags given * Allow Path object in every method that accept path as string * Some exceptions were renamed, some useless were removed ## \[6.1.2] - 2022-11-10 ### Fixed * Use of typing.List instead of list to support python < 3.9 ## \[6.1.1] - 2022-11-10 ### Added * Add convert\_tags\_to\_classification() in dataset version * Add get\_data\_tags() on asset ### Changed * Github workflow ## \[6.1.0] - 2022-11-01 ### Added * Add ONXX in Framework enums * Possibility to use strings instead of enums in methods ### Changed * Make duration optional in annotation creation/update/overwrite * Make duration int compatible * log() method now create, update or append a log, in only one call to backend * Remove create\_log method * Optional type hint is now used everywhere * Prevent user from updating log image via log.update(), only experiment.log can be used for this case ### Fixed * In log() method, image type data stored bad values ## \[6.0.2] - 2022-10-17 ### Added * Logging File for an Experiment ### Changed * A lot of typos, variable naming, minor formatting * String .format() into f-strings * Old package dependencies * train\_test\_split() return MultiAsset instead of List\[Asset] ### Fixed * Regeneration of JWT when expired * When downloading file, open file only if response is ok ## \[6.0.1] - 2022-10-04 ### Added * CHANGELOG file. ### Changed * Fixed test\_train\_split with breaking changes of query language in list assets. * Documentation of core functions, minor typos fixes. * Workflow for testing on staging when push on development and main. ## \[6.0.0] - 2022-10-03 ### Added * Annotation are now objects storing Rectangle, Polygon, Classification, Point and Line for an Asset * Artifact are now Experiment files stored * DatasetVersion are now used as versions of Dataset * Datalake objects * Datasource objects * Job objects, to wait for tasks to end * ModelVersion are now used as versions of Model * ScanFiles are now Scan files stored ### Changed * Pictures renamed to Asset that are Data objects in a DatasetVersion * Multithreading usage * Tags * Annotations with objects