Sparse R-CNN

open-mmlab / mmdetection

Last updated on Feb 23, 2021
Sparse R-CNN (R-101-FPN, 3x, pytorch, prop_no=100, MS=True)
lr sched 3x
Backbone Layers 101
File Size 478.48 MB
Training Data MS COCO
Training Resources 8x NVIDIA V100 GPUs
Training Time
Architecture Sparse R-CNN, ResNet, FPN
MS train True
lr sched 3x
Backbone Layers 101
Number of Proposals 100
SHOW MORE
SHOW LESS
Sparse R-CNN (R-101-FPN, 3x, pytorch, prop_no=300, MS=True)
lr sched 3x
Backbone Layers 101
File Size 478.68 MB
Training Data MS COCO
Training Resources 8x NVIDIA V100 GPUs
Training Time
Architecture Sparse R-CNN, ResNet, FPN
MS train True
lr sched 3x
Backbone Layers 101
Number of Proposals 300
SHOW MORE
SHOW LESS
Sparse R-CNN (R-50-FPN, 1x, pytorch, prop_no=100, MS=False)
lr sched 1x
Backbone Layers 50
File Size 405.78 MB
Training Data MS COCO
Training Resources 8x NVIDIA V100 GPUs
Training Time
Architecture Sparse R-CNN, ResNet, FPN
MS train False
lr sched 1x
Backbone Layers 50
Number of Proposals 100
SHOW MORE
SHOW LESS
Sparse R-CNN (R-50-FPN, 3x, pytorch, prop_no=100, MS=True)
lr sched 3x
Backbone Layers 50
File Size 405.78 MB
Training Data MS COCO
Training Resources 8x NVIDIA V100 GPUs
Training Time
Architecture Sparse R-CNN, ResNet, FPN
MS train True
lr sched 3x
Backbone Layers 50
Number of Proposals 100
SHOW MORE
SHOW LESS
Sparse R-CNN (R-50-FPN, 3x, pytorch, prop_no=300, MS=True)
lr sched 3x
Backbone Layers 50
File Size 405.98 MB
Training Data MS COCO
Training Resources 8x NVIDIA V100 GPUs
Training Time
Architecture Sparse R-CNN, ResNet, FPN
MS train True
lr sched 3x
Backbone Layers 50
Number of Proposals 300
SHOW MORE
SHOW LESS
README.md

Sparse R-CNN: End-to-End Object Detection with Learnable Proposals

Introduction

[ALGORITHM]

@article{peize2020sparse,
  title   =  {{SparseR-CNN}: End-to-End Object Detection with Learnable Proposals},
  author  =  {Peize Sun and Rufeng Zhang and Yi Jiang and Tao Kong and Chenfeng Xu and Wei Zhan and Masayoshi Tomizuka and Lei Li and Zehuan Yuan and Changhu Wang and Ping Luo},
  journal =  {arXiv preprint arXiv:2011.12450},
  year    =  {2020}
}

Results and Models

Model Backbone Style Lr schd Number of Proposals Multi-Scale RandomCrop box AP Config Download
Sparse R-CNN R-50-FPN pytorch 1x 100 False False 37.9 config model | log
Sparse R-CNN R-50-FPN pytorch 3x 100 True False 42.8 config model | log
Sparse R-CNN R-50-FPN pytorch 3x 300 True True 45.0 config model | log
Sparse R-CNN R-101-FPN pytorch 3x 100 True False 44.2 config model | log
Sparse R-CNN R-101-FPN pytorch 3x 300 True True 46.2 config model | log

Notes

We observe about 0.3 AP noise especially when using ResNet-101 as the backbone.

Results

Object Detection on COCO minival
MODEL BOX AP
Sparse R-CNN (R-101-FPN, 3x, pytorch, prop_no=300, MS=True) 46.2
Sparse R-CNN (R-50-FPN, 3x, pytorch, prop_no=300, MS=True) 45.0
Sparse R-CNN (R-101-FPN, 3x, pytorch, prop_no=100, MS=True) 44.2
Sparse R-CNN (R-50-FPN, 3x, pytorch, prop_no=100, MS=True) 42.8
Sparse R-CNN (R-50-FPN, 1x, pytorch, prop_no=100, MS=False) 37.9
Contact us on: hello@paperswithcode.com . Papers With Code is a free resource with all data licensed under CC-BY-SA.
Terms Data policy Cookies policy