GCNet for Object Detection

By Yue Cao, Jiarui Xu, Stephen Lin, Fangyun Wei, Han Hu.

This repo is a official implementation of "GCNet: Non-local Networks Meet Squeeze-Excitation Networks and Beyond" on COCO object detection based on open-mmlab's mmdetection. The core operator GC block could be find here. Many thanks to mmdetection for their simple and clean framework.

Update on 2020/12/07

The extension of GCNet got accepted by TPAMI (PDF).

Update on 2019/10/28

GCNet won the Best Paper Award at ICCV 2019 Neural Architects Workshop!

Update on 2019/07/01

The code is refactored. More results are provided and all configs could be found in configs/gcnet.

Notes: Both PyTorch official SyncBN and Apex SyncBN have some stability issues. During training, mAP may drops to zero and back to normal during last few epochs.

Update on 2019/06/03

GCNet is supported by the official mmdetection repo here. Thanks again for open-mmlab's work on open source projects.

Introduction

GCNet is initially described in arxiv. Via absorbing advantages of Non-Local Networks (NLNet) and Squeeze-Excitation Networks (SENet), GCNet provides a simple, fast and effective approach for global context modeling, which generally outperforms both NLNet and SENet on major benchmarks for various recognition tasks.

Citing GCNet

@article{cao2019GCNet,
  title={GCNet: Non-local Networks Meet Squeeze-Excitation Networks and Beyond},
  author={Cao, Yue and Xu, Jiarui and Lin, Stephen and Wei, Fangyun and Hu, Han},
  journal={arXiv preprint arXiv:1904.11492},
  year={2019}
}

Main Results

Results on R50-FPN with backbone (fixBN)

| Back-bone | Model | Back-bone Norm | Heads | Context | Lr schd | Mem (GB) | Train time (s/iter) | Inf time (fps) | box AP | mask AP | Download | |:---------:|:----------------:|:-------------:|:----------------:|:--------------:|:-------:|:--------:|:-------------------:|:--------------:|:------:|:-------:|:--------:| | R50-FPN | Mask | fixBN | 2fc (w/o BN) | - | 1x | 3.9 | 0.453 | 10.6 | 37.3 | 34.2 | model| | R50-FPN | Mask | fixBN | 2fc (w/o BN) | GC(c3-c5, r16) | 1x | 4.5 | 0.533 | 10.1 | 38.5 | 35.1 | model| | R50-FPN | Mask | fixBN | 2fc (w/o BN) | GC(c3-c5, r4) | 1x | 4.6 | 0.533 | 9.9 | 38.9 | 35.5 | model| | R50-FPN | Mask | fixBN | 2fc (w/o BN) | - | 2x | - | - | - | 38.2 | 34.9 | model| | R50-FPN | Mask | fixBN | 2fc (w/o BN) | GC(c3-c5, r16) | 2x | - | - | - | 39.7 | 36.1 | model| | R50-FPN | Mask | fixBN | 2fc (w/o BN) | GC(c3-c5, r4) | 2x | - | - | - | 40.0 | 36.2 | model|

Results on R50-FPN with backbone (syncBN)

| Back-bone | Model | Back-bone Norm | Heads | Context | Lr schd | Mem (GB) | Train time (s/iter) | Inf time (fps) | box AP | mask AP | Download | |:---------:|:----------------:|:-------------:|:----------------:|:--------------:|:-------:|:--------:|:-------------------:|:--------------:|:------:|:-------:|:--------:| | R50-FPN | Mask | SyncBN | 2fc (w/o BN) | - | 1x | 3.9 | 0.543 | 10.2 | 37.2 | 33.8 | model| | R50-FPN | Mask | SyncBN | 2fc (w/o BN) | GC(c3-c5, r16) | 1x | 4.5 | 0.547 | 9.9 | 39.4 | 35.7 | model| | R50-FPN | Mask | SyncBN | 2fc (w/o BN) | GC(c3-c5, r4) | 1x | 4.6 | 0.603 | 9.4 | 39.9 | 36.2 | model| | R50-FPN | Mask | SyncBN | 2fc (w/o BN) | - | 2x | 3.9 | 0.543 | 10.2 | 37.7 | 34.3 | model| | R50-FPN | Mask | SyncBN | 2fc (w/o BN) | GC(c3-c5, r16) | 2x | 4.5 | 0.547 | 9.9 | 39.7 | 36.0 | model| | R50-FPN | Mask | SyncBN | 2fc (w/o BN) | GC(c3-c5, r4) | 2x | 4.6 | 0.603 | 9.4 | 40.2 | 36.3 | model| | R50-FPN | Mask | SyncBN | 4conv1fc (SyncBN) | - | 1x | - | - | - | 38.8 | 34.6 | model| | R50-FPN | Mask | SyncBN | 4conv1fc (SyncBN) | GC(c3-c5, r16) | 1x | - | - | - | 41.0 | 36.5 | model| | R50-FPN | Mask | SyncBN | 4conv1fc (SyncBN) | GC(c3-c5, r4) | 1x | - | - | - | 41.4 | 37.0 | model|

Results on stronger backbones

| Back-bone | Model | Back-bone Norm | Heads | Context | Lr schd | Mem (GB) | Train time (s/iter) | Inf time (fps) | box AP | mask AP | Download | |:---------:|:----------------:|:-------------:|:----------------:|:--------------:|:-------:|:--------:|:-------------------:|:--------------:|:------:|:-------:|:--------:| | R101-FPN | Mask | fixBN | 2fc (w/o BN) | - | 1x | 5.8 | 0.571 | 9.5 | 39.4 | 35.9 | model| | R101-FPN | Mask | fixBN | 2fc (w/o BN) | GC(c3-c5, r16) | 1x | 7.0 | 0.731 | 8.6 | 40.8 | 37.0 | model| | R101-FPN | Mask | fixBN | 2fc (w/o BN) | GC(c3-c5, r4) | 1x | 7.1 | 0.747 | 8.6 | 40.8 | 36.9 | model| | R101-FPN | Mask | SyncBN | 2fc (w/o BN) | - | 1x | 5.8 | 0.665 | 9.2 | 39.8 | 36.0 | model| | R101-FPN | Mask | SyncBN | 2fc (w/o BN) | GC(c3-c5, r16) | 1x | 7.0 | 0.778 | 9.0 | 41.1 | 37.4 | model| | R101-FPN | Mask | SyncBN | 2fc (w/o BN) | GC(c3-c5, r4) | 1x | 7.1 | 0.786 | 8.9 | 41.7 | 37.6 | model| | X101-FPN | Mask | SyncBN | 2fc (w/o BN) | - | 1x | 7.1 | 0.912 | 8.5 | 41.2 | 37.3 | [model](htt