RuoyuChen10 / objectdetection-saliency-maps

Based on the mmdetection framework, compute various salience maps for object detection.

Object Detection Saliency Maps

Based on mmdetection framework. You need to install MMDetaction first, follow here: get_started.md

An installation example (cuda 11.6):

conda create -n detX python=3.9
conda activate detX
pip install torch==1.12.1+cu116 torchvision==0.13.1+cu116 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu116
pip install -v -e .
pip install mmcv-full==1.7.0 -f https://download.openmmlab.com/mmcv/dist/cu116/torch1.12/index.html

Update

• [2023.01.17] I released the Grad-CAM visualization results based on the single-stage object detection method, RetinaNet.

• [2023.01.16] I released the Grad-CAM visualization results based on the two-stage object detection method, Faster R-CNN.

1. Grad-CAM

Selvaraju, Ramprasaath R., et al. "Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization." International Journal of Computer Vision 128.2 (2020): 336-359.

Paper Url: https://arxiv.org/abs/1610.02391

Supported Object Detection Algorithm:

Yolo V3

Paper: https://arxiv.org/abs/1804.02767

Step by step see: gradcam-yolov3.ipynb

Model config and checkpoint: https://huggingface.co/RuoyuChen/objectdetection-saliency-maps/

python gradcam-yolov3.py \
        --config <Configs Path> \
        --checkpoint <Checkpoint Path> \
        --image-path <Your Image Path> \
        --bbox-index 0 \
        --save-dir images/GradCAM/YOLOV3

Visualization:

Faster R-CNN (C4)

Paper: https://arxiv.org/abs/1506.01497

Step by step see: gradcam-faster-rcnn-C4-proposal.ipynb and gradcam-faster-rcnn-C4-global.ipynb

mkdir checkpoints
cd checkpoints
wget https://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_caffe_c4_1x_coco/faster_rcnn_r50_caffe_c4_1x_coco_20220316_150152-3f885b85.pth
cd ..

visualization based on proposal:

python gradcam-frcn-c4-proposal.py \
        --config <Configs Path> \
        --checkpoint <Checkpoint Path> \
        --image-path <Your Image Path> \
        --bbox-index 0 \
        --save-dir images/GradCAM/FRCN-C4

visualization based on global:

python gradcam-frcn-c4-global.py \
        --config <Configs Path> \
        --checkpoint <Checkpoint Path> \
        --image-path <Your Image Path> \
        --bbox-index 0 \
        --save-dir images/GradCAM/FRCN-C4

RetinaNet

Paper: https://arxiv.org/abs/1708.02002

No step by step

wget https://download.openmmlab.com/mmdetection/v2.0/retinanet/retinanet_r50_fpn_1x_coco/retinanet_r50_fpn_1x_coco_20200130-c2398f9e.pth

python gradcam-retinanet.py \
        --config <Configs Path> \
        --checkpoint <Checkpoint Path> \
        --image-path <Your Image Path> \
        --bbox-index 0 \
        --save-dir images/GradCAM/RetinaNet

Visualization:

2. D-RISE

Vitali Petsiuk, Rajiv Jain, Varun Manjunatha, Vlad I. Morariu, Ashutosh Mehra, Vicente Ordonez, Kate Saenko; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021, pp. 11443-11452

Paper Url: https://openaccess.thecvf.com/content/CVPR2021/html/Petsiuk_Black-Box_Explanation_of_Object_Detectors_via_Saliency_Maps_CVPR_2021_paper.html

Supported Object Detection Algorithm:

Yolo V3

Paper: https://arxiv.org/abs/1804.02767

Step by step see: drise-yolov3.ipynb

Model config and checkpoint: https://huggingface.co/RuoyuChen/objectdetection-saliency-maps/

python drise-yolov3.py \
        --config <Configs Path> \
        --checkpoint <Checkpoint Path> \
        --image-path <Your Image Path> \
        --bbox-index 0 \
        --save-dir images/DRISE/YOLOV3

Visualization: