3.7.1

This GitHub Release was done automatically by CircleCI

3.7.0: updated __version__ (#1943)

This GitHub Release was done automatically by CircleCI

3.6.1

New Features

• Added DistributedSamplerWrapper to automatically wrap non-dist samplers in cases we use dist mode by @NatanBagrov in #1856
• YoloNAS_Pose_Fine_Tuning_Animals_Pose_Dataset by @ofrimasad in #1876
• Introduce fp16 flag to enable/disable mixed precision for predict() by @BloodAxe in #1881
• Feature/sg 1386 granular control over export in ptq and qat by @BloodAxe in #1879

Deprecations

• Deprecate tight_box_rotation parameters in COCODetectionDataset by @BloodAxe in #1786

Improvements

• pycocotools dependency removed by @BloodAxe in #1791
• Added explicit antialias=False to ensure we can export torchvision Resize to ONNX by @BloodAxe in #1824
• Updated colab notebook to include line that initializes plugins by @BloodAxe in #1822
• Feature/sg 000 add note that qat only supports gpu by @shaydeci in #1830
• Added more hydra resolvers by @NatanBagrov in #1829
• Added crash tip for the case when SGLogger is None by @shaydeci in #1799
• Added YoloNAS-Pose fine-tuning notebook by @BloodAxe in #1831
• allow flexibility to provide absolute path to annotations by @NatanBagrov in #1840
• fix vulnerabilities by @ofrimasad in #1861
• fix onnx version by @ofrimasad in #1863
• fix tarfile extraction by @ofrimasad in #1868

Bugfixes

• fixed an issue with eval forcing to have a val_dataloader in config by @NatanBagrov in #1823
• Fix typo error in ann_areas vs ann_area attribute by @BloodAxe in #1828
• Added fixed random seed to not depend of randomness of initialized weights by @BloodAxe in #1839
• Fixed a wrong color channel order when processing images from webcamera and improved exception message when on MacOS by @BloodAxe in #1821
• Bugfix by @ofrimasad in #1874
• fix a bug when ploting a dataset with images in a range other than 0-255 by @ofrimasad in #1884
• Fixed speed of COCO dataset parsing by @BloodAxe in #1888

Other

• Update welcome.md by @ofrimasad in #1790
• Update README.md - voxel51 integration by @Shani-Perl in #1827
• fix vulnerability by @ofrimasad in #1872
• Bug/sg 1247 reoarganize tests by @shaydeci in #1789
• Added warning message for dataset license by @shaydeci in #1846

Full Changelog: 3.6.0...3.6.1

3.6.0

Hey @channel
We're pleased to announce the release of Super-Gradients 3.6.0 🙂 .
This update includes several important changes and improvements:
Changes and Enhancements

• Added segmentation samples and support for albumentation transforms for segmentation, contributed by @shaydeci.
• Implemented distance-based detection matching in DetectionMetrics as an enhancement by @DimaBir.
• New training hyperparameter - finetune, and multiple LR assignment by @shaydeci, read about it here
• Enhanced ImagePermute processing inclusion, by @BloodAxe.
• Improved dataset plotting and plot functionality, by
  @Louis Dupont
  .
• Updated prediction notebooks and documentation, thanks to
  @Louis Dupont
  .
• Supported depth estimation dataset and added depth estimation metrics support, by @shaydeci.
• Proposed an API for checking model input compatibility, by @BloodAxe.
• Extended predict() support for segmentation models, by @BloodAxe.
• Removed deprecated features from version 3.6.0, by @shaydeci.
• Updated pre-trained models badge URL, contributed by @gasparitiago.
• Made changes to PPYoloELoss, removing the requirement for a reg_max parameter, by @BloodAxe.
• Switched to using onnxsim instead of onnx-simplifier for consistency in naming, thanks to @BloodAxe.
  Bugfixes
• Resolved a bug in OhemLoss thanks to @danielafrimi.
• Updated conditions to ensure functionality only on rank 0 where [context.sg](http://context.sg/)_logger is available, by @shaydeci.
• Modified the default set_device value to prevent unintentional launch of DDP, updated by
  @Louis Dupont
  .
• Addressed a bug where multigpu=None with device=cpu wasn't functioning as expected, thanks to
  @Louis Dupont
  .
• Adjusted bounding box thickness and text size relative to bbox size, for object detection model's predict() by
  @Louis Dupont
  .
• Addressed a bug in DetectionMixup that affected YoloXTrainingStageSwitchCallback, by @BloodAxe.
• Corrected a typo in an exception message variable name, by @BloodAxe.
• Reintegrated tests and refined CI/CD workflows, thanks to @shaydeci and @Yonatan-Kaplounov.
• Fixed import issues and improved model flexibility and metrics handling, mainly by @BloodAxe.
• Ensured class names in DetectionDataset are contained within a trivial container, by @BloodAxe.
• Fixed ExtremeBatchDetectionVisualizationCallback for multiscale collate function, by @BloodAxe.
• Several bug fixes and improvements in DistanceBasedDetectionMetrics and DetectionMetrics, by @BloodAxe.
  And various other fixes and improvements across the board to enhance functionality and user experience.
  For a detailed list of changes, refer to the full changelog.
  New Contributors
• Welcome to the contributors' team, @DimaBir, @gasparitiago, and @Yonatan-Kaplounov, for making their first contributions.Hey @channel
  We're pleased to announce the release of Super-Gradients 3.6.0 🙂 .
  This update includes several important changes and improvements:
  Changes and Enhancements
• Added segmentation samples and support for albumentation transforms for segmentation, contributed by @shaydeci.
• Implemented distance-based detection matching in DetectionMetrics as an enhancement by @DimaBir.
• New training hyperparameter - finetune, and multiple LR assignment by @shaydeci, read about it here
• Enhanced ImagePermute processing inclusion, by @BloodAxe.
• Improved dataset plotting and plot functionality, by
  @Louis Dupont
  .
• Updated prediction notebooks and documentation, thanks to
  @Louis Dupont
  .
• Supported depth estimation dataset and added depth estimation metrics support, by @shaydeci.
• Proposed an API for checking model input compatibility, by @BloodAxe.
• Extended predict() support for segmentation models, by @BloodAxe.
• Removed deprecated features from version 3.6.0, by @shaydeci.
• Updated pre-trained models badge URL, contributed by @gasparitiago.
• Made changes to PPYoloELoss, removing the requirement for a reg_max parameter, by @BloodAxe.
• Switched to using onnxsim instead of onnx-simplifier for consistency in naming, thanks to @BloodAxe.
  Bugfixes
• Resolved a bug in OhemLoss thanks to @danielafrimi.
• Updated conditions to ensure functionality only on rank 0 where [context.sg](http://context.sg/)_logger is available, by @shaydeci.
• Modified the default set_device value to prevent unintentional launch of DDP, updated by
  @Louis Dupont
  .
• Addressed a bug where multigpu=None with device=cpu wasn't functioning as expected, thanks to
  @Louis Dupont
  .
• Adjusted bounding box thickness and text size relative to bbox size, for object detection model's predict() by
  @Louis Dupont
  .
• Addressed a bug in DetectionMixup that affected YoloXTrainingStageSwitchCallback, by @BloodAxe.
• Corrected a typo in an exception message variable name, by @BloodAxe.
• Reintegrated tests and refined CI/CD workflows, thanks to @shaydeci and @Yonatan-Kaplounov.
• Fixed import issues and improved model flexibility and metrics handling, mainly by @BloodAxe.
• Ensured class names in DetectionDataset are contained within a trivial container, by @BloodAxe.
• Fixed ExtremeBatchDetectionVisualizationCallback for multiscale collate function, by @BloodAxe.
• Several bug fixes and improvements in DistanceBasedDetectionMetrics and DetectionMetrics, by @BloodAxe.
  And various other fixes and improvements across the board to enhance functionality and user experience.
  For a detailed list of changes, refer to the full changelog.
  New Contributors
• Welcome to the contributors' team, @DimaBir, @gasparitiago, and @Yonatan-Kaplounov, for making their first contributions..

3.5.0

This GitHub Release was done automatically by CircleCI

New features

• A model.predict(video) does not cause OOM anymore by @hakuryuu96 in #1621
• Learning rates can now be specified per-layer (Can now freeze layers or train backbone with lower LR) by @shaydeci in #1612
• A model.predict can now take skip_resize argument to run forward in the original image resolution (Good for large images & small objects) by @Louis-Dupont in #1605
• Added support of multiple test loaders in train_from_config by @BloodAxe in #1641

Documentation

• Added documentation for DatasetAdapter by @Louis-Dupont in #1591
• Added notebook for adding a custom dataset by @Louis-Dupont in #1592
• Added notebooks on segmentation (quick start & transfer learning) by @shaydeci in #1634
• Add notebook for fine-tuning YoloNAS with QAT by @BloodAxe in #1638

Bugfixes

• Fixed bug in quantized YoloNAS model that let do degraded performance of the exported model by @BloodAxe in #1638
• Fix numpy deprecation warning on creating ragged array when using _pad_image by @BloodAxe in #1632
• Fixed bug in model.export() that led to crash when FP16 export was requested and model was on CPU device by @BloodAxe in #1643
• Fix Invalid syntax at convert_recipe_to_code.py by @seunghalee1226 in #1642
• Fixed bug that lead to incorrect visualization of target bboxes when using DetectionVisualization.visualize_batch by @BloodAxe in #1652
• Fixed bug in Data Adapter for Segmentation task by @Louis-Dupont in #1654

New Contributors

• @seunghalee1226 made their first contribution in #1642

Full Changelog: 3.4.1...3.5.0

3.4.1

Bugfixes

• Fix serialization of ListConfig in checkpoint state dictionary by @BloodAxe in #1534
• Ensure that checkpoint_num_classes is propagated from YAML to model by @BloodAxe in #1533
• Fixed a bug in YoloNASPose.export() that prevented to export model for BS>1 by @BloodAxe in #1530
• Check class_id validity in DetectionDataset by @Louis-Dupont in #1536
• Fix training_params deprecation by @Louis-Dupont in #1542
• Fixing typos and adding extra info to CONTRIBUTING by @hakuryuu96 in #1546
• Remove np.bool which is not supported in latest np versions by @Louis-Dupont in #1558
• Fix CSPDarknet53.foward by @Louis-Dupont in #1564
• Bugfix of model.export() to work correct with bs>1 by @BloodAxe in #1551
• Mixed precision automatically changed with warning by @Louis-Dupont in #1567
• Updated data-gradients requirement by @shaydeci in #1572
• Fixed issue with torch 1.12 where _scale_fn_ref is missing in CyclicLR by @BloodAxe in #1575
• Fixed issue with torch 1.12 issue with arange not supporting fp16 for CPU device. by @BloodAxe in #1574
• Reorder operators to ensure Neg operator is not used by @BloodAxe in #1584
• Updated "What are Recipes and How To Use Them" notebook by @BloodAxe in #1586
• Update link for "Open in Colab" by @BloodAxe in #1588
• Added metrics logging to checkpoint and separate yaml file by @hakuryuu96 in #1562
• Fixed bug in ModelWeightAveraging class that led to corrupted model when metric to watch was NaN/Inf by @BloodAxe in #1598
• Transfer learning classification notebook update by @shaydeci in #1587
• Update KD Notebook for classification by @BloodAxe in #1595
• Fixed bug in COCOPoseEstimationDataset which caused the images or reversed channel order used in predict() by @BloodAxe in #1609
• Fixed supervisely .get which uses dataset: inside the .yaml file by @shaydeci in #1603
• Fixed bug in _pad_image that did not support pad_value=(R,B,G) input by @BloodAxe in #1599
• Fixed ListConfig in pose estimation dataset classes by @BloodAxe in #1602
• Enforced check of that all notebooks has matching SG version installed by @BloodAxe in #1607
• Update README by @BloodAxe in #1615
• Way to fix bug with validation frequency by @hakuryuu96 in #1601
• Changed key for SegKDLoss by @hakuryuu96 in #1620
• Bump onnx-simplifier version require at least 0.4.3 by @BloodAxe in #1631

Enhancement:

• Add class_names to model.predict for detection by @Louis-Dupont in #1529
• Argument max_batches support to training log and tqdm progress bar. by @hakuryuu96 in #1554
• Add explanations on experiment management by @Louis-Dupont in #1559
• Added recipe unifying script by @shaydeci in #1560
• Added packages to pth checkpoint by @hakuryuu96 in #1581
• Added sequential assignment capability to PPYoloELoss by @BloodAxe in #1582
• Add replace in channels by @Louis-Dupont in #1557
• Add image examples by @Louis-Dupont in #1589
• Update ptq and qat training by @BloodAxe in #1618
• Added convert_from_recipe script, that converts yaml train_from_config, "recipe style" training script to a "python dictionaries" standalone training script. by @BloodAxe in #1568

New Contributors

• @hakuryuu96 made their first contribution in #1546
• @aler9 made their first contribution in #1548

3.4.0

New features

• Added convert_recipe_to_code script to convert YAML recipe to self-contained train script by @BloodAxe in #1568
• Added export_recipe script to convert a YAML recipe to a single big YAML file by @shaydeci in #1560
• YoloNAS-Pose by @BloodAxe in #1611

Improvements

• Added sequential assignment capability to PPYoloELoss by @BloodAxe in #1582
• Add class_names to model.predict for detection by @Louis-Dupont in #1529
• Check class_id validity in DetectionDataset by @Louis-Dupont in #1536
• [Improvement] max_batches support to training log and tqdm progress bar. by @hakuryuu96 in #1554
• Reorder operators to ensure Neg operator is not used by @BloodAxe in #1584
• Feature/sg 1198 mixed precision automatically changed with warning by @Louis-Dupont in #1567
• Feature/sg 1053 add explanations on experiment managing by @Louis-Dupont in #1559
• Feature/sg 849 add replace in channels by @Louis-Dupont in #1557
• Updated "What are Recipes and How To Use Them" notebook by @BloodAxe in #1586
• Added metrics logging to checkpoint and separate yaml file by @hakuryuu96 in #1562
• Added packages to pth checkpoint by @hakuryuu96 in #1581

Bugfixes

• Fix serialization of ListConfig in checkpoint state dictionary by @BloodAxe in #1534
• Fixed a bug in YoloNASPose.export() that prevented to export model for BS>1 by @BloodAxe in #1530
• Ensure that checkpoint_num_classes is propagated from YAML to model by @BloodAxe in #1533
• Bugfix of model.export() to work correct with bs>1 by @BloodAxe in #1551
• Fixed bug in COCOPoseEstimationDataset which caused the images or reversed channel order used in predict() by @BloodAxe in #1609
• Fixed bug in ModelWeightAveraging class that led to corrupted model when metric to watch was NaN/Inf by @BloodAxe in #1598
• Fixed issue with torch 1.12 where _scale_fn_ref is missing in CyclicLR by @BloodAxe in #1575
• Fixed issue with torch 1.12 issue with arange not supporting fp16 for CPU device. by @BloodAxe in #1574
• fixed supervisely .get which uses dataset: inside the .yaml file by @shaydeci in #1603
• Fixed bug in _pad_image that did not support pad_value=(R,B,G) input by @BloodAxe in #1599
• Fix CSPDarknet53.foward by @Louis-Dupont in #1564

Other

• Dependency of data-gradients version bumped up to 0.2.2
• Remove np.bool which is not supported in latest np versions by @Louis-Dupont in #1558
• Enforced check of that all notebooks has matching SG version installed by @BloodAxe in #1607
• Fixing typos and adding extra info to CONTRIBUTING by @hakuryuu96 in #1546
• Fix typo in class documentation by @aler9 in #1548

What's Changed

• Fix training_params deprecation by @Louis-Dupont in #1542
• Fixed incorrect automatic variable used by @BloodAxe in #1565
• Update link for "Open in Colab" by @BloodAxe in #1588
• Feature/sg 1190 transfer learning cls notebook by @shaydeci in #1587
• Update KD Notebook for classification by @BloodAxe in #1595

New Contributors

• @hakuryuu96 made their first contribution in #1546
• @aler9 made their first contribution in #1548

Full Changelog: 3.3.0...3.4.0

3.3.1

Super Gradients 3.3.1

Improvements

• Add class_names to model.predict for detection by @Louis-Dupont in #1529
• Data-Gradients dependency bumped to 0.2.2 by @Louis-Dupont in #1541
• Check class_id validity in DetectionDataset by @Louis-Dupont in #1536
• Fixing typos and adding extra info to CONTRIBUTING by @hakuryuu96 in #1546
• Progress bar during training now respects max_train_batches/max_valid_batches by @hakuryuu96 #1554

Bugfixes

• ListConfig/DictConfig types do not leak into checkpoint state dictionary anymore by @BloodAxe in #1534
• Migrate usage of np.bool -> bool which is not supported in latest np versions by @Louis-Dupont in #1558
• Ensure that checkpoint_num_classes is propagated from YAML files to models.get by @BloodAxe in #1533
• Fixed detection models export to ONNX bug for batch size > 1 @BloodAxe in #1530
• Fix CSPDarknet53.foward by @Louis-Dupont in #1564
• Fixed incorrect automatic variable used by @BloodAxe in #1565
• Fix typo in class documentation by @aler9 in #1548
• Trainer does not crash anymore when using CPU and Automated Mixed Precision is enabled in training params @Louis-Dupont in #1567
• Fixed issue with torch 1.12 where _scale_fn_ref is missing in CyclicLR by @BloodAxe in #1575
• Fixed issue with torch 1.12 issue with arange not supporting fp16 for CPU device. by @BloodAxe in #1574

New Contributors

• @hakuryuu96 made their first contribution in #1546
• @aler9 made their first contribution in #1548

Full Changelog: 3.3.0...3.3.1

3.3.0

This GitHub Release was done automatically by CircleCI

What's Changed

• Update yolox_loss.py by @eran-deci in #1265
• Fix documentation link! by @Louis-Dupont in #1300
• Addiing more logs to let user know when pretrained_weights is being used, and/or downloaded by @Louis-Dupont in #1298
• set HYDRA_FULL_ERROR by @Louis-Dupont in #1291
• remove arch_params from logs by @Louis-Dupont in #1297
• Make classification models inherit from BaseClassifier by @Louis-Dupont in #1314
• Make cache annotation optional by @Louis-Dupont in #1332
• Fix class_inclusion_list in DetectionDataset by @Louis-Dupont in #1327
• Support for segmentation extreme batch cases by @shaydeci in #1282
• Feature/sg 901 extreme batch visualization for object detection by @shaydeci in #1339
• Update OD docs with clarified output formats by @BloodAxe in #1348
• Fix model.get by @Louis-Dupont in #1287
• Feature/sg 1059 fix ci by @BloodAxe in #1350
• Fix datasetparams not showing by @Louis-Dupont in #1317
• Fix a bug in implementation of DetectionRGB2BGR.get_equivalent_preprocessing by @BloodAxe in #1352
• Add utilities to plot datasets to Weights & Biases + Add callback to log validation predictions to Weights & Biases by @soumik12345 in #1167
• DagsHub Logger: Fix unsupported metric formats for MLflow, Add example notebook by @nirbarazida in #915
• New Export API by @BloodAxe in #1318
• Feature/sg 000 fix predict in pose estimation by @BloodAxe in #1358
• Add model export tutorial to documentation website by @Louis-Dupont in #1362
• fix by @Louis-Dupont in #1367
• Feature/sg 000 propagate imagenet dataset params by @BloodAxe in #1368
• Doc changes by @pranoyr in #1253
• Summarize models, losses & metrics for segmentation by @BloodAxe in #1354
• Feature/sg 000 fix import of onnx graphsurgeon by @BloodAxe in #1359
• Feature/sg 1047 predict od with labels by @shaydeci in #1365
• Feature/sg 1033 fix yolox anchors by @BloodAxe in #1369
• version bumped by @shaydeci in #1374
• Adding pose estimation to the readme by @BloodAxe in #1375
• Remove broken link that we can't recover where it was pointing to by @BloodAxe in #1376
• Segformer swi signed by @Yael-Baron in #1361
• added instructions for extracting bounding boxes by @AkashParua in #1312
• [Adding Fix to Master] Explicitly remove the stride keys from the checkpoint by @Louis-Dupont in #1397
• PLFM-3938 Create a code-loading utility in DeciClient by @roikoren755 in #1393
• closing inactive issues by @bit-scientist in #1398
• Feature/sg 1027 checkpoint directory refacto by @Louis-Dupont in #1401
• Update doc by @Louis-Dupont in #1381
• Feature/sg 1057 add breaking change tests by @Louis-Dupont in #1373
• Move device only when needed in predict by @Louis-Dupont in #1394
• Feature/sg 1039 add factory doc by @Louis-Dupont in #1395
• Fix checkpoint loading with run_id by @Louis-Dupont in #1403
• fix bug in convert from recipe by @ofrimasad in #1406
• Feature/sg 907 add yaml explanation by @Louis-Dupont in #1422
• Feature/sg 1039 improve recipe explanation by @Louis-Dupont in #1404
• Added docs on LR & DDP by @BloodAxe in #1414
• Update the docs to reflect the variable_setup by @BloodAxe in #1424
• Hotfix/sg 000 improve breaking change detection with imports by @Louis-Dupont in #1420
• Adding docs to clarify required properties in recipe file by @BloodAxe in #1429
• Only test breaking changes in src + ignoring examples by @Louis-Dupont in #1430
• Add deprecate module by @Louis-Dupont in #1416
• Fix DDP checkpoint by @Louis-Dupont in #1415
• fix regression tests by @ofrimasad in #1408
• Moving changes from master_320 to master by @BloodAxe in #1428
• Improve Exception when metric_to_watch is wrong by @Louis-Dupont in #1437
• Change release filter rule to support rc tagging master_320 and other release branches (#1436) by @BloodAxe in #1438
• Fix invalid syntax at YOLONAS.md by @Pslydhh in #1441
• Adding a deprecation option into registeries by @Louis-Dupont in #1421
• Fix checkpoint doc by @Louis-Dupont in #1445
• Updating OD example by adding additional details on customizing metri… by @BloodAxe in #1449
• Correctly handle Ctrl-C event by @BloodAxe in #1461
• Update Segmentation.md by @donatoaz in #1464
• Fix Breaking Change Detection on Inheritence by @Louis-Dupont in #1458
• Hotfix/sg 000 fix predict loading from np torch by @Louis-Dupont in #1419
• Feature/sg 1041 rename object names by @Louis-Dupont in #1446
• fixed registry for @register_loss(LabelSmoothingCrossEntropyLoss) by @shaydeci in #1475
• Fixed loading preprocessing params from pretrained weights by @BloodAxe in #1473
• bit-scientist-update-qat_ptq_yolo_nas.md by @bit-scientist in #1440
• Added instructions what to do when PR contains unsigned commits by @BloodAxe in #1486
• Fixed bug of no saving simplified ONNX file by @BloodAxe in #1489
• Redirect logs based on env var by @jnccd in #1474
• Fix error with pip _get_installed_distributions by @Louis-Dupont in #1494
• Remove target for Detection CollateFN by @Louis-Dupont in #1470
• Improve extreme batch visualization callbacks by @BloodAxe in #1488
• Feature/sg 1147 notebooks in repo by @BloodAxe in #1493
• A call to _process_collate_fn_params was missing in dataloaders.get. by @BloodAxe in #1501
• Hotfix/sg 000 fix sigmas by @BloodAxe in #1503
• tests skipped by @shaydeci in #1509
• Feature/sg 1144 new export in quantization by @BloodAxe in #1511
• Small improvements to utils by @BloodAxe in #1513
• ExtremeBatchCaseVisualizationCallback now has access to additional_batch_items by @BloodAxe in #1517
• fixed cyclic lr state dict by @shaydeci in #1469
• Introduce sample-centric keypoint transforms by @BloodAxe in #1492
• Fix YoloNAS on cuda by @Louis-Dupont in #1444
• Feature/sg 1172 criterion params removal by @shaydeci in #1519
• Squashed changes with YoloNASPose & Loss by @BloodAxe in #1512
• Feature/sg 000 add adapter by @Louis-Dupont in https://github.com/Deci-AI/su...

3.2.1

3.2 1 - Minor bugfixes release

TLDR:

• Improvements in docs 📜
• A few fixes in export API without introducing breaking changes 💪

What's Changed

• Update yolox_loss.py by @eran-deci in #1265
• Fix documentation link! by @Louis-Dupont in #1300
• Addiing more logs to let user know when pretrained_weights is being used, and/or downloaded by @Louis-Dupont in #1298
• set HYDRA_FULL_ERROR by @Louis-Dupont in #1291
• remove arch_params from logs by @Louis-Dupont in #1297
• Make classification models inherit from BaseClassifier by @Louis-Dupont in #1314
• Make cache annotation optional by @Louis-Dupont in #1332
• Fix class_inclusion_list in DetectionDataset by @Louis-Dupont in #1327
• Support for segmentation extreme batch cases by @shaydeci in #1282
• Feature/sg 901 extreme batch visualization for object detection by @shaydeci in #1339
• Update OD docs with clarified output formats by @BloodAxe in #1348
• Fix model.get by @Louis-Dupont in #1287
• Feature/sg 1059 fix ci by @BloodAxe in #1350
• Fix datasetparams not showing by @Louis-Dupont in #1317
• Fix a bug in implementation of DetectionRGB2BGR.get_equivalent_preprocessing by @BloodAxe in #1352
• Add utilities to plot datasets to Weights & Biases + Add callback to log validation predictions to Weights & Biases by @soumik12345 in #1167
• DagsHub Logger: Fix unsupported metric formats for MLflow, Add example notebook by @nirbarazida in #915
• New Export API by @BloodAxe in #1318
• Feature/sg 000 fix predict in pose estimation by @BloodAxe in #1358
• Add model export tutorial to documentation website by @Louis-Dupont in #1362
• fix by @Louis-Dupont in #1367
• Feature/sg 000 propagate imagenet dataset params by @BloodAxe in #1368
• Doc changes by @pranoyr in #1253
• Summarize models, losses & metrics for segmentation by @BloodAxe in #1354
• Feature/sg 000 fix import of onnx graphsurgeon by @BloodAxe in #1359
• Feature/sg 1047 predict od with labels by @shaydeci in #1365
• Feature/sg 1033 fix yolox anchors by @BloodAxe in #1369
• version bumped by @shaydeci in #1374
• Bugfix: Fix loading DeciDet models that has anchors keys in state dict. by @BloodAxe in #1386
• BaseKeypointsDataset now inherits from HasPreprocessingParams by @BloodAxe in #1380
• Hotfix/sg 000 fix yolox replace head by @BloodAxe in #1411
• Export API fix: Raise meaningful exception if model has no preprocessing metadata but preprocessing=True by @BloodAxe in #1413
• Feature/sg 1106 Minor fixes in export api by @BloodAxe in #1412
• Hotfix/sg 000 Fix support of arbitrary number of heads by @BloodAxe in #1431
• Version bump to 3.2.1 by @BloodAxe in #1435
• Change release filter rule to support rc tagging master_320 and other release branches by @BloodAxe in #1436
• Hardcode release tag filter to master_320 by @BloodAxe in #1448

New Contributors

• @nirbarazida made their first contribution in #915
• @pranoyr made their first contribution in #1253

Full Changelog: 3.1.3...3.2.1