ZF-Yu / Sewer-defect-detection

Composite transformer multi-stage defect detection for sewer pipes

Demo

Environments

• windows 10
• GPU RTX8000×1
• cuda 11.1
• cudnn 8
• python 3.7
• pytorch 1.8.1
• mmcv-full 1.4.0
• yapf 0.32.0
• albumentations 1.3.0
• apex 0.1 (optional)

Installation

pip install -v -e .

Offline data augmentation

Brightness adjustment and Sharpen offline data augmentation.

python projects\dataaug.py 

Model train

It is recommended that the dataset be converted to an annotation file in coco format.

#Automatic mixed-precision training
python tools\train.py projects\cbnet\cbswinl_cascade.py <input_images_path> <path_of_modified _contrast_images> <path_of_offline_images>
#Without Automatic mixed-precision training
python tools\train.py projects\cbnet\cbswinl_cascade_woamp.py <input_images_path> <path_of_modified _contrast_images> <path_of_offline_images>

Model inference

#With out TTA
python tools\test.py projects\cbnet\cbswinl_cascade.py work_dirs\cbswinl_cascade\latest.pth --eval bbox
#With TTA
python tools\test.py projects\cbnet\cbswinl_cascade_tta.py work_dirs\cbswinl_cascade\latest.pth --eval bbox

Model box fusion

Note the paths to the prediction file on line 19 and the result file on line 89 of wbf.py. The final file is wbf_result.json.

python projects\wbf.py