Robust and Accurate Object Detection via Self-Knowledge Distillation

Last update: Jul 01, 2022

Related tags

Deep Learning udfa

Overview

Robust and Accurate Object Detection via Self-Knowledge Distillation

paper:https://arxiv.org/abs/2111.07239

Environments

Python 3.7
Cuda 10.1
Prepare dependency

Notes: We revise MMCV to adapt adversarial algorithms. Therefore we suggest that you prepare environments strictly as follows:

conda create -n udfa python=3.7
conda activate udfa
sh prepare_env.sh

Prepare datasets

VOC0712, download from http://host.robots.ox.ac.uk/pascal/VOC/, and place it under data directory
COCO2017, download from https://cocodataset.org/#download, and place it under data directory
The structure of datasets is shown as follows:

Train

VOC dataset

Generate GFLV2-R34 pretrained detector (served as teacher) on PASCAL_VOC 0712:

python -m torch.distributed.launch --nproc_per_node=4  train.py --launcher pytorch --config configs/gflv2/gflv2_r34_fpn_voc_std.py 
cd work_dirs/gflv2_r34_fpn_voc_std
cp epoch_12.pth ../../weights/gflv2_r34_voc_pre.pth

Training GFLV2-R34 using udfa on PASCAL_VOC 0712:

python -m torch.distributed.launch --nproc_per_node=4  train.py --launcher pytorch --config configs/gflv2/gflv2_r34_fpn_voc_kdss.py --load-from weights/gflv2_r34_voc_pre.pth

Training GFLV2-R34 using udfa with advprop on PASCAL_VOC 0712:

python -m torch.distributed.launch --nproc_per_node=4  train.py --launcher pytorch --config configs/gflv2/gflv2_r34_fpn_voc_kdss1.py --load-from weights/gflv2_r34_voc_pre.pth

Training GFLV2-R34 using Det-AdvProp on PASCAL_VOC 0712:

python -m torch.distributed.launch --nproc_per_node=4  train.py --launcher pytorch --config configs/gflv2/gflv2_r34_fpn_voc_mixbn.py --load-from weights/gflv2_r34_voc_pre.pth

COCO dataset

Generate GFLV2-R34 pretrained detector (served as teacher) on COCO:

python -m torch.distributed.launch --nproc_per_node=4  train.py --launcher pytorch --config configs/gflv2/gflv2_r34_fpn_coco_std.py 
cd work_dirs/gflv2_r34_fpn_coco_std
cp epoch_12.pth ../../weights/gflv2_r34_coco_pre.pth

Training GFLV2-R34 using udfa on COCO:

python -m torch.distributed.launch --nproc_per_node=4  train.py --launcher pytorch --config configs/gflv2/gflv2_r34_fpn_coco_kdss.py --load-from weights/gflv2_r34_coco_pre.pth

Training GFLV2-R34 using Det-AdvProp on COCO:

python -m torch.distributed.launch --nproc_per_node=4  train.py --launcher pytorch --config configs/gflv2/gflv2_r34_fpn_coco_mixbn.py --load-from weights/gflv2_r34_coco_pre.pth

Test

Evlauate the clean AP or adversarial robustness on PASCAL_VOC 2007 test set:

python -m torch.distributed.launch --nproc_per_node=4 test.py --launcher pytorch --configs/gflv2/gflv2_r34_fpn_voc_std.py  --checkpoint weights/gflv2_r34_voc_pre.pth --num_steps 0 --step_size 2 --eval mAP

Evlauate the clean AP or adversarial robustness on COCO 2017val set:

python -m torch.distributed.launch --nproc_per_node=4 test.py --launcher pytorch --configs/gflv2/gflv2_r34_fpn_coco_std.py  --checkpoint weights/gflv2_r34_coco_pre.pth --num_steps 0 --step_size 2 --eval bbox

Acknowledgement

Our project is based on ImageCorruptions, MMDetection and MMCV.

Robust and Accurate Object Detection via Self-Knowledge Distillation

Related tags

Overview

Robust and Accurate Object Detection via Self-Knowledge Distillation

Environments

Prepare datasets

Train

VOC dataset

COCO dataset

Test

Acknowledgement

Owner

Weipeng Xu

IEEE Winter Conference on Applications of Computer Vision 2022 Accepted

The official repository for Deep Image Matting with Flexible Guidance Input

Torchlight2 lan game server tool - A message forwarding tool for Torchlight 2 lan game

A Pytorch Implementation of Domain adaptation of object detector using scissor-like networks

Deformable DETR is an efficient and fast-converging end-to-end object detector.

A python comtrade load library accelerated by go

A Python module for the generation and training of an entry-level feedforward neural network.

Implementation of H-UCRL Algorithm

Code for: Imagine by Reasoning: A Reasoning-Based Implicit Semantic Data Augmentation for Long-Tailed Classification

Learning To Have An Ear For Face Super-Resolution

Animation of solving the traveling salesman problem to optimality using mixed-integer programming and iteratively eliminating sub tours

Hippocampal segmentation using the UNet network for each axis

pytorch, hand(object) detect ,yolo v5，手检测

This is an official implementation for the WTW Dataset in "Parsing Table Structures in the Wild " on table detection and table structure recognition.

Implementation of Transformer in Transformer, pixel level attention paired with patch level attention for image classification, in Pytorch

BigbrotherBENL - Face recognition on the Big Brother episodes in Belgium and the Netherlands.

Sharpened cosine similarity torch - A Sharpened Cosine Similarity layer for PyTorch

A large dataset of 100k Google Satellite and matching Map images, resembling pix2pix's Google Maps dataset.

A Deep Learning Framework for Neural Derivative Hedging

A python-image-classification web application project, written in Python and served through the Flask Microframework. This Project implements the VGG16 covolutional neural network, through Keras and Tensorflow wrappers, to make predictions on uploaded images.