abhigoku10 / Yolact_minimal

Minimal PyTorch implementation of YOLACT.

Yolact_minimal

Minimal PyTorch implementation of Yolact:《YOLACT: Real-time Instance Segmentation》.
The original project is here.

This implementation simplified the original code, preserved the main function and made the network easy to understand.

Following instruction is based on resnet-101.

The network structure.

Environments

PyTorch 1.1.
Python 3.6 or above.
Other common packages.

Prepare

• Firstly, modify the paths of training and val datasets in data/config.py in line 72-77.
• If some directories are missed, just create them by yourself.
• Download the pretrained weights.
  ImageNet pretrained weights
  Yolact pretrained weights

Train

# Trains using the base config with a batch size of 8 (the default).
python train.py --config=yolact_base_config

# Resume training (just pass the .pth file to the model by using --resume).
python train.py --config=yolact_base_config --resume weights/yolact_base_2_35000.pth

# Using --batch_size, --lr, --momentum, --decay to set the batch size, learning rate, momentum and weight decay.
python train.py --config=yolact_base_config --batch_size=4

Val

# Evaluate on COCO val2017.
python eval.py --trained_model=weights/yolact_base_54_800000.pth

The results should be:

# Create a json file and then use the COCO API to evaluate the result.
python eval.py --config=yolact_base_config --cocoapi
# Then,
python coco_eval.py

# Benchmark
python eval.py --trained_model=weights/yolact_base_54_800000.pth --benchmark --max_images=1000

Detect

# Detect images, pass the path of your image directory to --image.
python detect.py --trained_model=weights/yolact_base_54_800000.pth --image images
# Detect a video, pass the path of your video to --video.
python detect.py --trained_model=weights/yolact_base_54_800000.pth --video video/1.mp4
# Use --hide_mask, --hide_score, --show_lincomb and so on to get different results.
python detect.py --trained_model=weights/yolact_base_54_800000.pth --image images --hide_mask