open-mmlab/mmdetection v2.13.0

MMDetection V2.13.0 Release

on GitHub

Highlights

• Support new methods: CenterNet, Seesaw Loss, MobileNetV2

New Features

• Support paper Objects as Points (#4602)
• Support paper Seesaw Loss for Long-Tailed Instance Segmentation (CVPR 2021) (#5128)
• Support MobileNetV2 backbone and inverted residual block (#5122)
• Support MIM (#5143)
• ONNX exportation with dynamic shapes of CornerNet (#5136)
• Add mask_soft config option to allow non-binary masks (#4615)
• Add PWC metafile (#5135)

Bug Fixes

• Fix YOLOv3 FP16 training error (#5172)
• Fix Cascade R-CNN TTA test error when det_bboxes length is 0 (#5221)
• Fix iou_thr variable naming errors in VOC recall calculation function (#5195)
• Fix Faster R-CNN performance dropped in ONNX Runtime (#5197)
• Fix DETR dict changed error when using python 3.8 during iteration (#5226)

Improvements

• Refactor ONNX export of two-stage detector (#5205)
• Replace MMDetection's EvalHook with MMCV's EvalHook for consistency (#4806)
• Update RoI extractor for ONNX (#5194)
• Use better parameter initialization in YOLOv3 head for higher performance (#5181)
• Release new DCN models of Mask R-CNN by mixed-precision training (#5201)
• Update YOLOv3 model weights (#5229)
• Add DetectoRS ResNet-101 model weights (#4960)
• Discard boxes with sizes equals to min_bbox_size (#5011)
• Remove duplicated code in DETR head (#5129)
• Remove unnecessary object in the class definition (#5180)

New Contributors

• @namihagi made their first contribution in #5121
• @zlx-6 made their first contribution in #5129
• @haofanwang made their first contribution in #5147
• @ijjustinKim made their first contribution in #5163
• @liuhuiCNN made their first contribution in #5192
• @SemyonBevzuk made their first contribution in #5197

Full Changelog: v2.12.0...v2.13.0