Training Extensions Versions Save

What's Changed

• Bump datumaro version by @yunchu in https://github.com/openvinotoolkit/training_extensions/pull/2502
• Upgrade nncf version for 1.4 release by @sovrasov in https://github.com/openvinotoolkit/training_extensions/pull/2459
• Update version for release 1.4.2rc5 by @yunchu in https://github.com/openvinotoolkit/training_extensions/pull/2507

Full Changelog: https://github.com/openvinotoolkit/training_extensions/compare/1.4.2rc4...1.4.2rc5

What's Changed

• Bump datumaro version to 1.5.0rc0 by @yunchu in https://github.com/openvinotoolkit/training_extensions/pull/2470
• Set tox version constraint by @yunchu in https://github.com/openvinotoolkit/training_extensions/pull/2472
• Bug fix for albumentations by @ashwinvaidya17 in https://github.com/openvinotoolkit/training_extensions/pull/2467
• Add a dummy hierarchical config required by MAPI by @sovrasov in https://github.com/openvinotoolkit/training_extensions/pull/2483

Full Changelog: https://github.com/openvinotoolkit/training_extensions/compare/1.4.2rc2...1.4.2rc4

What's Changed

• Update PTQ ignored scope for hrnet 18 mod2 by @sovrasov in https://github.com/openvinotoolkit/training_extensions/pull/2449
• Fix OpenVINO inference for legacy models by @ashwinvaidya17 in https://github.com/openvinotoolkit/training_extensions/pull/2450
• Update for 1.4.2rc2 by @yunchu in https://github.com/openvinotoolkit/training_extensions/pull/2455
• Prevent zero-sized saliency map in tiling if tile size is too big by @sovrasov in https://github.com/openvinotoolkit/training_extensions/pull/2452
• Update pot fq reference number by @yunchu in https://github.com/openvinotoolkit/training_extensions/pull/2456

Full Changelog: https://github.com/openvinotoolkit/training_extensions/compare/1.4.2rc1...1.4.2rc2

What's Changed

• Add workaround for the incorrect meta info M-RCNN (used for XAI) by @negvet in https://github.com/openvinotoolkit/training_extensions/pull/2437
• Add model category attributes to model template (for releases/1.4.0) by @goodsong81 in https://github.com/openvinotoolkit/training_extensions/pull/2439
• Update for 1.4.2rc1 by @yunchu in https://github.com/openvinotoolkit/training_extensions/pull/2441
• Fix label list order for h-label classification by @GalyaZalesskaya in https://github.com/openvinotoolkit/training_extensions/pull/2440
• Modified fq numbers for lite HRNET by @yunchu in https://github.com/openvinotoolkit/training_extensions/pull/2445

Full Changelog: https://github.com/openvinotoolkit/training_extensions/compare/1.4.1...1.4.2rc1

Enhancements

• Update the README file in exportable code (https://github.com/openvinotoolkit/training_extensions/pull/2411)

Bug fixes

• Fix broken links in documentation (https://github.com/openvinotoolkit/training_extensions/pull/2405)

Full Changelog: https://github.com/openvinotoolkit/training_extensions/compare/1.4.0...1.4.1

New features

• Support encrypted dataset training (https://github.com/openvinotoolkit/training_extensions/pull/2209)
• Add custom max iou assigner to prevent CPU OOM when large annotations are used (https://github.com/openvinotoolkit/training_extensions/pull/2228)
• Auto train type detection for Semi-SL, Self-SL and Incremental: "--train-type" now is optional (https://github.com/openvinotoolkit/training_extensions/pull/2195)
• Add per-class XAI saliency maps for Mask R-CNN model (https://github.com/openvinotoolkit/training_extensions/pull/2227)
• Add new object detector Deformable DETR (https://github.com/openvinotoolkit/training_extensions/pull/2249)
• Add new object detector DINO (https://github.com/openvinotoolkit/training_extensions/pull/2266)
• Add new visual prompting task (https://github.com/openvinotoolkit/training_extensions/pull/2203, https://github.com/openvinotoolkit/training_extensions/pull/2274, https://github.com/openvinotoolkit/training_extensions/pull/2311, https://github.com/openvinotoolkit/training_extensions/pull/2354, https://github.com/openvinotoolkit/training_extensions/pull/2318)
• Add new object detector ResNeXt101-ATSS (https://github.com/openvinotoolkit/training_extensions/pull/2309)

Enhancements

• Introduce channel_last parameter to improve the performance (https://github.com/openvinotoolkit/training_extensions/pull/2205)
• Decrease time for making a workspace (https://github.com/openvinotoolkit/training_extensions/pull/2223)
• Set persistent_workers and pin_memory as True in detection task (https://github.com/openvinotoolkit/training_extensions/pull/2224)
• New algorithm for Semi-SL semantic segmentation based on metric learning via class prototypes (https://github.com/openvinotoolkit/training_extensions/pull/2156)
• Self-SL for classification now can recieve just folder with any images to start contrastive pretraining (https://github.com/openvinotoolkit/training_extensions/pull/2219)
• Update OpenVINO version to 2023.0, and NNCF verion to 2.5 (https://github.com/openvinotoolkit/training_extensions/pull/2090)
• Improve XAI saliency map generation for tiling detection and tiling instance segmentation (https://github.com/openvinotoolkit/training_extensions/pull/2240)
• Remove CenterCrop from Classification test pipeline and editing missing docs link(https://github.com/openvinotoolkit/training_extensions/pull/2375)
• Switch to PTQ for sseg (https://github.com/openvinotoolkit/training_extensions/pull/2374)

Bug fixes

• Fix the bug that auto adapt batch size is unavailable with IterBasedRunner (https://github.com/openvinotoolkit/training_extensions/pull/2182)
• Fix the bug that learning rate isn't scaled when multi-GPU trianing is enabled(https://github.com/openvinotoolkit/training_extensions/pull/2254)
• Fix the bug that label order is misaligned when model is deployed from Geti (https://github.com/openvinotoolkit/training_extensions/pull/2369)
• Fix NNCF training on CPU (https://github.com/openvinotoolkit/training_extensions/pull/2373)
• Fix H-label classification (https://github.com/openvinotoolkit/training_extensions/pull/2377)
• Fix invalid import structures in otx.api (https://github.com/openvinotoolkit/training_extensions/pull/2383)
• Add for async inference calculating saliency maps from predictions (Mask RCNN IR) (https://github.com/openvinotoolkit/training_extensions/pull/2395)

Known issues

• OpenVINO(==2023.0) IR inference is not working well on 2-stage models (e.g. Mask-RCNN) exported from torch==1.13.1

New Contributors

• @Wovchena made their first contribution in https://github.com/openvinotoolkit/training_extensions/pull/1952
• @QSmally made their first contribution in https://github.com/openvinotoolkit/training_extensions/pull/2046

Full Changelog: https://github.com/openvinotoolkit/training_extensions/compare/v1.0.0...1.4.0

What's Changed

• Fix invalid import structures in otx.api by @harimkang in https://github.com/openvinotoolkit/training_extensions/pull/2383
• Update for 1.4.0rc4 by @yunchu in https://github.com/openvinotoolkit/training_extensions/pull/2385

Full Changelog: https://github.com/openvinotoolkit/training_extensions/compare/1.4.0rc3...1.4.0rc4

What's Changed

• Fix NNCF training on CPU by @sovrasov in https://github.com/openvinotoolkit/training_extensions/pull/2373
• Align label order between Geti and OTX by @eunwoosh in https://github.com/openvinotoolkit/training_extensions/pull/2369
• Remove CenterCrop from Classification test pipeline and editing missing docs link by @sungmanc in https://github.com/openvinotoolkit/training_extensions/pull/2375
• Fix H-label classification by @sungmanc in https://github.com/openvinotoolkit/training_extensions/pull/2377
• Update for release 1.4 by @yunchu in https://github.com/openvinotoolkit/training_extensions/pull/2380
• Switch to PTQ for sseg by @sovrasov in https://github.com/openvinotoolkit/training_extensions/pull/2374

Full Changelog: https://github.com/openvinotoolkit/training_extensions/compare/1.4.0rc2...1.4.0rc3

What's Changed

• Skip openvino integration test by @sungchul2 in https://github.com/openvinotoolkit/training_extensions/pull/2323
• Fix import dm.DatasetSubset by @vinnamkim in https://github.com/openvinotoolkit/training_extensions/pull/2324
• Fix semantic segmentation soft prediction dtype by @negvet in https://github.com/openvinotoolkit/training_extensions/pull/2322
• Contrain yapf verison lesser than 0.40.0 by @eunwoosh in https://github.com/openvinotoolkit/training_extensions/pull/2328
• Fix detection e2e tests by @jaegukhyun in https://github.com/openvinotoolkit/training_extensions/pull/2327
• Mergeback: Label addtion/deletion 1.2.4 --> 1.4.0 by @sungmanc in https://github.com/openvinotoolkit/training_extensions/pull/2326
• Bump datumaro up to 1.4.0rc2 by @yunchu in https://github.com/openvinotoolkit/training_extensions/pull/2332
• Tiling Doc for releases 1.4.0 by @eugene123tw in https://github.com/openvinotoolkit/training_extensions/pull/2333
• Bump otx version to 1.4.0rc2 by @yunchu in https://github.com/openvinotoolkit/training_extensions/pull/2341
• Add OTX deploy for visual prompting task by @sungchul2 in https://github.com/openvinotoolkit/training_extensions/pull/2311
• Bump albumentations version in anomaly requirements by @djdameln in https://github.com/openvinotoolkit/training_extensions/pull/2350
• Update action detection by @jaegukhyun in https://github.com/openvinotoolkit/training_extensions/pull/2346
• Fix e2e by @jaegukhyun in https://github.com/openvinotoolkit/training_extensions/pull/2348
• Update ModelAPI in 1.4 release by @sovrasov in https://github.com/openvinotoolkit/training_extensions/pull/2347
• Add OTX optimize for visual prompting task by @sungchul2 in https://github.com/openvinotoolkit/training_extensions/pull/2318
• Update detection docs by @jaegukhyun in https://github.com/openvinotoolkit/training_extensions/pull/2335
• Add visual prompting documentation by @sungchul2 in https://github.com/openvinotoolkit/training_extensions/pull/2354
• Remove custom modelapi patch in visual prompting by @sungchul2 in https://github.com/openvinotoolkit/training_extensions/pull/2359
• Fix graph metric order and label issues by @goodsong81 in https://github.com/openvinotoolkit/training_extensions/pull/2356
• Update multi-label document and conversion script by @sungmanc in https://github.com/openvinotoolkit/training_extensions/pull/2358
• Update third party programs by @yunchu in https://github.com/openvinotoolkit/training_extensions/pull/2365
• Make anomaly task compatible with older albumentations versions by @djdameln in https://github.com/openvinotoolkit/training_extensions/pull/2363
• Fixing detection saliency map for one class case by @negvet in https://github.com/openvinotoolkit/training_extensions/pull/2368
• Add e2e test for visual prompting by @sungchul2 in https://github.com/openvinotoolkit/training_extensions/pull/2360
• Fix e2e by @jaegukhyun in https://github.com/openvinotoolkit/training_extensions/pull/2366
• Add Dino head unit tests by @jaegukhyun in https://github.com/openvinotoolkit/training_extensions/pull/2344
• Update for release 1.4.0rc2 by @yunchu in https://github.com/openvinotoolkit/training_extensions/pull/2370

Full Changelog: https://github.com/openvinotoolkit/training_extensions/compare/1.4.0rc1...1.4.0rc2

Bug fixes

• Fix 'None' node issue in label schema mapping in case of label deletion (https://github.com/openvinotoolkit/training_extensions/pull/2300)
• Squeeze anomaly mpas in metadata (https://github.com/openvinotoolkit/training_extensions/pull/2310)
• Fix semantic segmentation soft prediction dtype (https://github.com/openvinotoolkit/training_extensions/pull/2321)
• Fix label deletion issue (https://github.com/openvinotoolkit/training_extensions/pull/2315)

Enhancements

• Per-class saliency maps for M-RCNN (https://github.com/openvinotoolkit/training_extensions/pull/2301)
• Disable semantic segmentation soft prediction processing (https://github.com/openvinotoolkit/training_extensions/pull/2302)
• Update export and nncf hyperparameters (https://github.com/openvinotoolkit/training_extensions/pull/2306)

Full Changelog: https://github.com/openvinotoolkit/training_extensions/compare/1.2.3.5...1.2.4