Explainable AI (XAI) aims to make machine learning models transparent by providing clear, understandable explanations of their decisions. This enhances trust, accountability, and the ability to debug and improve AI systems.
The implemented workflow architecture of our research work is published & available online Download here Link
#Results -Training-batch
HAM10000 Dataset
| Model | AK | BCC | BKL | DF | MEL | NV | SCC | VASC |
|---|---|---|---|---|---|---|---|---|
| YOLOv7-XAI | 98.9 | 97.4 | 96.5 | 96.7 | 97.4 | 96.0 | 96.4 | 95.0 |
| YOLOv7 | 95.8 | 94.9 | 96.0 | 94.1 | 94.2 | 92.3 | 95.2 | 94.5 |
| YOLOv6 | 94.8 | 95.3 | 94.4 | 93.9 | 93.4 | 90.8 | 94.4 | 93.0 |
Docker environment (recommended)
Expand
# create the docker container, you can change the share memory size if you have more.
nvidia-docker run --name YOLOv7-XAI -it -v your_coco_path/:/coco/ -v your_code_path/:/YOLOv7-XAI --shm-size=64g nvcr.io/nvidia/pytorch:21.08-py3
# apt install required packages
apt update
apt install -y zip htop screen libgl1-mesa-glx
# pip install required packages
pip install seaborn thop
# go to code folder
cd /YOLOv7-XAIpython test.py --data data/data.yaml --img 640 --batch 32 --conf 0.001 --iou 0.65 --device 0 --weights YOLOv7-XAI.pt --name YOLOv7-XAI640_valTo measure accuracy, download
Data preparation
labelimg annotation tool
- Download MS HAM10000 dataset images HAM10000-annotations for Pycocotools (email:v.nirmalaresearch@gmail.com) for annotated images and label folder sharing
Single GPU training
# train p5 models
python train.py --workers 8 --device 0 --batch-size 32 --data data/data.yaml --img 640 640 --cfg cfg/training/YOLOv7-XAI.yaml --weights '' --name YOLOv7-XAI --hyp data/hyp.scratch.p5.yaml
# train p6 models
python train_aux.py --workers 8 --device 0 --batch-size 16 --data data/data.yaml --img 1280 1280 --cfg cfg/training/YOLOv7-XAI-w6.yaml --weights '' --name YOLOv7-XAI-w6 --hyp data/hyp.scratch.p6.yamlMultiple GPU training
# train p5 models
python -m torch.distributed.launch --nproc_per_node 4 --master_port 9527 train.py --workers 8 --device 0,1,2,3 --sync-bn --batch-size 128 --data data/data.yaml --img 640 640 --cfg cfg/training/YOLOv7-XAI.yaml --weights '' --name YOLOv7-XAI --hyp data/hyp.scratch.p5.yaml
# train p6 models
python -m torch.distributed.launch --nproc_per_node 8 --master_port 9527 train_aux.py --workers 8 --device 0,1,2,3,4,5,6,7 --sync-bn --batch-size 128 --data data/data.yaml --img 1280 1280 --cfg cfg/training/YOLOv7-XAI-w6.yaml --weights '' --name YOLOv7-XAI-w6 --hyp data/hyp.scratch.p6.yamlOn image:
python detect.py --weights YOLOv7-XAI.pt --conf 0.25 --img-size 640 --source inference/images/ISIC12399.jpg
Pytorch to CoreML (and inference on MacOS/iOS)
Pytorch to ONNX with NMS (and inference)
python export.py --weights YOLOv7-XAI-tiny.pt --grid --end2end --simplify \
--topk-all 100 --iou-thres 0.65 --conf-thres 0.35 --img-size 640 640 --max-wh 640Pytorch to TensorRT with NMS (and inference)
wget https://github.com/Nirmala-research/YOLOv7-XAI/releases/download/v0.1/YOLOv7-XAI-tiny.pt
python export.py --weights ./YOLOv7-XAI-tiny.pt --grid --end2end --simplify --topk-all 100 --iou-thres 0.65 --conf-thres 0.35 --img-size 640 640
git clone https://github.com/Linaom1214/tensorrt-python.git
python ./tensorrt-python/export.py -o YOLOv7-XAI-tiny.onnx -e YOLOv7-XAI-tiny-nms.trt -p fp16Pytorch to TensorRT another way
Expand
wget https://github.com/Nirmala-research/YOLOv7-XAI/releases/download/v0.1/YOLOv7-XAI-tiny.pt
python export.py --weights YOLOv7-XAI-tiny.pt --grid --include-nms
git clone https://github.com/Linaom1214/tensorrt-python.git
python ./tensorrt-python/export.py -o YOLOv7-XAI-tiny.onnx -e YOLOv7-XAI-tiny-nms.trt -p fp16
# Or use trtexec to convert ONNX to TensorRT engine
/usr/src/tensorrt/bin/trtexec --onnx=YOLOv7-XAI-tiny.onnx --saveEngine=YOLOv7-XAI-tiny-nms.trt --fp16Tested with: Python 3.7.13, Pytorch 1.12.0+cu113
