MMCV Release V1.2.3

New Features

• Add custom operators support for onnxruntime. (#612)
• Support mpi launcher. (#726)

Improvements

• Add 'iof' mode and 'aligned' mode for box_iou_rotated. (#726, #753)
• Refactor mmcv-full installation to fit pip>=20.3 and follow PEP440. (#748)
• Add model links (MobileNet-V3, ResNet18-V1c). (#755, #757)
• Allow to replace nested tuple and list via options in Config. (#731)
• Support mmcv bricks in flops compuation. (#715)

Bug Fixes

• Fix mmcls related models and links. (#741, #745)
• Fix grad_output uncontiguous. (#737)
• Avoid recursion error when using customized norm layer without _abbr_. (#732)
• Minor bug fixes, docs, unit tests related to nms. (#728, #733)
• Fix checkpoint key error. (#722)
• Fix comments of init_weights in ConvModule (#730)

MMCV Release V1.2.2

New Features

• Add clamp activation layer. (#685)

Improvements

• Refactor flow_warp and remove its C implementation. (#707)
• LossScaler: Allow the initial value to be set in dynamic mode. (#692)
• Add installation instructions for mmcv-full build on Windows. (#663)
• Allow list index keys in Config.merge_from_dict. (#696)

Bug Fixes

• Check MASTER_ADDR in the environment variable before assignment. (#705)
• Fix mmcv-full installation using pip 20.3. (#703)

MMCV Release V1.2.1

New Features

• Add box_iou_rotated, ml_nms_rotated and nms_rotated. (#625)
• Add wrappers for Conv3d, MaxPool3d and ConvTranspose3d. (#652)
• Add onnx support to roi_align and roi_pool. (#593)
• Add more transforms: LUT(#646) and CLAHE(#647).

Improvements

• Better support of pytorch2onnx for detection models. (#648)
• Add documentation for registry. (#656)
• Refactor logger. (#659)
• Enhance hsigmoid: support more parameters. (#657)

Bug Fixes

• Fix rotated NMS. (#668)

MMCV Release V1.2.0

New Features

• Support PyTorch 1.7. (#631)
• Add dynamic scale that manages loss scaling in mixed precision training. (#585)

Improvements

• Add hook message for checkpoint hook (#635)

Bug Fixes

• Fix bug in logger hooks. (#634, #637)
• Update lr_updater to correct period index computation. (#574)
• Fix missing arguments when converting DCN to ONNX. (#624)

MMCV Release V1.1.6

New Features

• New method import_modules_from_strings for import customized modules when loading config from file. (#606)
• Add switch for onnx exporter to get different ONNX model according to the backend. (#607)
• Add Sync buffer in CheckpointHook. (#588)

Improvements

• Refactor logger hooks. (#605)
• Skip opencv requirement if it's already installed in the env. (#616)
• Serveral updates regarding Windows and unittests. (#586)

Bug Fixes

• Fix bug in resuming dict version of optimizers when using epoch based runner. (#610)
• Fix iter-time-hook bug when meeting iter-based runner. (#623)

MMCV Release V1.1.5

Bug Fixes

• Fix wrappers version comparison bug. (#602)
• Revert DCN to fix the bug introduced in #565. (#597)

MMCV Release V1.1.4

New Features

• Add a RUNNERS registry and build_runner method. (#570)
• Support to specify LR of DCN's conv_offset. (#344)

Improvements

• merge the calling of train/val_step and batch_processor into run_iter (#553)
• Add the missing by_epoch arg for MlflowLoggerHook and WandbLoggerHook. (#576)
• Add aligned=True for torchvision roi_align. (#581)
• Adopt the new URL https://download.openmmlab.com. (#583)
• Register deconv in CONV_LAYERS. (#582)
• Log grad norm in fp16 optimizer hook. (#584)

Bug Fixes

• Fix fp16_optimizer state bug. (#580)
• Fix the pytorch version comparison in op wrappers. (#572)

MMCV Release V1.1.3

New Features

• Implement Conv2dAdaptivePadding, which is the same op as 2D convolution in tensorflow with padding set to "same". (#529)
• Support reading video from url (#531)
• Add imtranslate, imequalize, adjust_color, adjust_brightness, adjust_contrast augmentation. (#538, #542, #543, #546)

Improvements

• Add switch for onnx exporter. (#564)
• Allow type to be default_args in build_from_cfg. (#558)
• Expose wrap_fp16_model to mmcv.runner. (#555)
• Move op wrappers into bricks and use wrappers in registry (#550)
• Support multiple interpolation modes for imrotate (#545)

Bug Fixes

• Fix the init method of deprecated wrappers. (#567)
• Fix DCN bug in parrots. (#562, #563, #565)
• Use copy instead of symlink on windows. (#557)
• Remove assertion for batch size and samples_per_gpu. (#549)
• Fix compilation bug in windows and add instruction for windows. (#540)
• Fix missing pad_val in impad_to_multiple. (#539)

MMCV V1.1.2 Release

New Features

• Add Swish activation (#522)
• Add shear augmentation (#526)
• Support loading checkpoints in MMClassification with the scheme mmcls:// (#511)

Improvements

• Allow imshow_det_bboxes() to return the image with bboxes. (#527)
• Add more default types for json dump (#536)

MMCV V1.1.1 Release

New Features

• Add collect_env() to collect environment info.
• Add import_modules_from_strings() to import modules from a string list.

Improvements

• Support splitting in batched_nms when box number is too large.
• Support pickling Config.
• Add diagonal flip mode.

Bug fixes

• Fix regex bugs in windows paths.
• Fix windows compilation issue.
• Fix inconsistent return types in pybind11 function prototypes.