v1.25.1

1.25.1 (2023-04-16)

Bug Fixes

• display Product Opener response during OFF product update if an (91d05d4)

v1.25.0

1.25.0 (2023-04-16)

Features

• add a function to send image to OFF (f6ac894)
• implement real multi-platform support (OFF, OBF,...) (9464f46)
• support multiple MongoDB DB (multi-project) (2c36b6f)

Bug Fixes

• add server_type field to logo indexed in ES (506ab02)
• add DISABLE_PRODUCT_CHECK settings (a201fae)
• fix in insert_images.py script (8dd914f)
• fix issue in settings (DISABLE_PRODUCT_CHECK value) (232e5c6)
• fix refresh-insight scheduled job (d363cf8)
• fix value for ENABLE_PRODUCT_CHECK in local env (20a58d5)
• fix wrong call to run_nutriscore_object_detection (a211249)
• fix wrongly formatted logging message (3ef4a0d)
• rename DISABLE_PRODUCT_CHECK into ENABLE_PRODUCT_CHECK (8c07478)
• rename en:gluten-free into en:no-gluten (e1f6417)
• replace call to lru_cache() by call to cache() (191faab)
• suppress mypy warnings (49a29b9)
• switch log level to DEBUG (d3ebd85)

Documentation

• add missing docstring parameter descriptions (32cc8f8)
• improve docstrings (d4bb80b)
• improve documentation in add-predictor.md (195f1fd)
• improve documentation in add-predictor.md (90bfda2)
• improve Robotoff API documentation (d930592)

v1.24.2

1.24.2 (2023-04-06)

Bug Fixes

• fix error in product weight insight (6866739)
• load lazily all resources in Robotoff (4dfa93f)
• move LogoLabelType to robotoff.types (0cc7efe)

Documentation

• add documentation about how to add a predictor (d467a4c)
• add documentation about interaction with Product Opener (05c0781)
• add references to codebase in category-prediction.md (56ccfba)

v1.24.1

1.24.1 (2023-04-05)

Bug Fixes

• fix incorrect offset in get_words_from_indices (547a867)

v1.24.0

1.24.0 (2023-04-05)

Features

• add function to get match bounding box (5d7eafa)
• allow to match text on OCRResult (73ae0e6)
• save bounding box information in OCR/flashtext predictions (9d4d432)

Bug Fixes

• add functions to delete/unselect an image (ade0294)
• allow partial match in get_words_from_indices (34cbd24)
• allow to match across blocks (c073129)
• bug fix in product weight insight generation (031e117)
• cache result word string in Word (30bc347)
• convert absolute coordinates to relative ones (e86e671)
• fix offset bug (a6e1e51)
• make error message easier to understand during HTTP 404 during OCR fetch (b352387)
• remove text_annotations OCRField and use new text field (f0637a4)
• save mapping between position of words and full annotation text (674ad77)
• use new computed text field in regex matching (b29e1db)
• use re.I flag instead of lowercasing string (0ecef5e)
• use strip_accents_v1 when necessary (f2ee677)

Documentation

• improve OCR class documentation (b461669)
• improve OCR documentation (0a76f83)

v1.23.1

1.23.1 (2023-03-16)

Bug Fixes

• always select deepest categorized nodes in category importer (aba33f8)
• improve healthcheck status check messages (e24e136)

v1.23.0

1.23.0 (2023-03-15)

Features

• add missing_category campaign to track products without categories (b103e66)
• cache image embeddings in DB (ImageEmbedding table) (ba25c75)
• store neighbor categories for v3 categorizer models (8d1d727)
• use keras new v3 model as default to predict categories (c0f55cf)
• use keras v3 model as default (a2d23a3)
• use keras_image_embeddings_3_0 by default in categorize CLI (737716e)

Bug Fixes

• allow to specify LOG_LEVEL in .env file (50a3aa4)
• deprecate campaign parameter in /questions* (82c26fe)
• exclude some categories from predictions (503fa8d)
• fix bug in save_image_embeddings (a600502)
• fix bug that occurs when image are missing in images table (1968f56)
• fix edge-case bug when no image is available (fac1ab1)
• fix newly introduced bug in category importer (a1bb507)
• fix serialization bug in predict (1a69ab2)
• fix SonarCloud-detected bug in BaseURLProvider (290a3fb)
• fix unit tests (61c6177)
• fix unit tests (0a3702f)
• ignore predicted category if it no longer exist in taxonomy (c3c4fbe)
• move save_image function to new robotoff.images module (6efdaf2)
• pass stub as argument in predict for easier testing (807c157)
• relax checks in save_images function (bcf998f)
• remove legacy unit tests (6f3e5e2)

Documentation

• add comment in build_triton_request function (cf1e8db)
• add documentation about category prediction (0a21476)

v1.22.1

1.22.1 (2023-03-13)

Bug Fixes

• don't generate image embedding/fetch OCR texts if not required (93fc96d)

v1.22.0

1.22.0 (2023-03-12)

Features

• add model with image embeddings as input (d79bbc2)

Bug Fixes

• add authentication for .net Product Opener when fetching products (18cbb20)
• add object detection label assets to repository (c925558)
• add support for git LFS (d6db888)
• disable cat matcher (en) for partial matches (649c016)
• don't keep numpy ndarray in debug.inputs dict (64e7124)
• fix integration test (b2b4f5a)
• increase CLIP max_batch_size to 32 (a39d61a)
• refactor category predictor data structure (812a406)
• update code after code review #1061 (deb20d8)
• update PUT /images/logos/LOGO_ID route to accept null value field (2e488ab)

Documentation

• add docstring (b4ace04)
• add documentation about install of git lfs (1cd9866)
• add documentation in metrics.py (0ce2b0e)

v1.21.0

1.21.0 (2023-02-28)

Features

• add new category classification models (60d167d)
• return debug information in /predict/category route (5b2b392)

Bug Fixes

• fix category predictions (347c72c)
• remove unused functions (5b0f1ae)
• silence false-positive mypy error (337777d)
• update /predict/category schema (f86c098)
• update poetry lock file (48d921c)

Documentation

• improve documentation in v3 category predictor code (675e056)