Skip to content

Sugar-Coder/Yolact_minimal

 
 

Repository files navigation

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.
This implementation has not been updated to Yolact++.

The network structure.

Example 0

Environments

PyTorch >= 1.1
Python >= 3.6
tensooardX
Other common packages.

Prepare

  • Build cython-nms
python setup.py build_ext --inplace
  • Download COCO 2017 datasets, modify self.data_root in 'res101_coco' in config.py.
  • Download weights.

Yolact trained weights.

Backbone box mAP mask mAP Google Drive Baidu Cloud
Resnet50 31.5 29.3 res50_coco_800000.pth password: 2v8s
Resnet101 32.9 30.5 res101_coco_800000.pth password: 0jq2

ImageNet pre-trained weights.

Backbone Google Drive Baidu Cloud
Resnet50 resnet50-19c8e357.pth password: a6ee
Resnet101 resnet101_reducedfc.pth password: kdht

Improvement log

2020.11.2. Improved data augmentation, use rectangle anchors, training is stable, infinite loss no longer appears.
2020.11.2. DDP training, train batch size increased to 16, +0.4 box mAP, +0.7 mask mAP (resnet101).

Train

# Train with resnet101 backbone on one GPU with a batch size of 8 (default).
python -m torch.distributed.launch --nproc_per_node=1 --master_port=$((RANDOM)) train.py --train_bs=8
# Train on multiple GPUs (i.e. two GPUs, 8 images per GPU).
export CUDA_VISIBLE_DEVICES=0,1  # Select the GPU to use.
python -m torch.distributed.launch --nproc_per_node=2 --master_port=$((RANDOM)) train.py --train_bs=16
# Train with other configurations (res101_coco, res50_coco, res50_pascal, res101_custom, res50_custom, in total).
python -m torch.distributed.launch --nproc_per_node=1 --master_port=$((RANDOM)) train.py --cfg=res50_coco
# Train with different batch_size (batch size should not be smaller than 4).
python -m torch.distributed.launch --nproc_per_node=1 --master_port=$((RANDOM)) train.py --train_bs=4
# Train with different image size (anchor settings related to image size will be adjusted automatically).
python -m torch.distributed.launch --nproc_per_node=1 --master_port=$((RANDOM)) train.py --img_size=400
# Resume training with the latest trained model.
python -m torch.distributed.launch --nproc_per_node=1 --master_port=$((RANDOM)) train.py --resume latest
# Resume training with a specified model.
python -m torch.distributed.launch --nproc_per_node=1 --master_port=$((RANDOM)) train.py --resume latest_res101_coco_35000.pth
# Set evalution interval during training, set -1 to disable it.  
python -m torch.distributed.launch --nproc_per_node=1 --master_port=$((RANDOM)) train.py --val_interval 8000
# Train on CPU.
python train.py --train_bs=4

Use tensorboard

tensorboard --logdir=tensorboard_log/res101_coco

Evalution

# Select the GPU to use.
export CUDA_VISIBLE_DEVICES=0
# Evaluate on COCO val2017 (configuration will be parsed according to the model name).
# The metric API in this project can not get the exact COCO mAP, but the evaluation speed is fast. 
python eval.py --weight=weights/res101_coco_800000.pth
# To get the exact COCO mAP:
python eval.py --weight=weights/res101_coco_800000.pth --coco_api
# Evaluate with a specified number of images.
python eval.py --weight=weights/res101_coco_800000.pth --val_num=1000
# Evaluate with traditional nms.
python eval.py --weight=weights/res101_coco_800000.pth --traditional_nms

Detect

  • detect result
    Example 2
# Select the GPU to use.
export CUDA_VISIBLE_DEVICES=0
# To detect images, pass the path of the image folder, detected images will be saved in `results/images`.
python detect.py --weight=weights/res101_coco_800000.pth --image=images
  • cutout object
    Example 3
# Use --cutout to cut out detected objects.
python detect.py --weight=weights/res101_coco_800000.pth --image=images --cutout
# To detect videos, pass the path of video, detected video will be saved in `results/videos`:
python detect.py --weight=weights/res101_coco_800000.pth --video=videos/1.mp4
# Use --real_time to detect real-timely.
python detect.py --weight=weights/res101_coco_800000.pth --video=videos/1.mp4 --real_time
  • linear combination result
    Example 4
# Use --hide_mask, --hide_score, --save_lincomb, --no_crop and so on to get different results.
python detect.py --weight=weights/res101_coco_800000.pth --image=images --save_lincomb

Train on PASCAL_SBD datasets

  • Download PASCAL_SBD datasets from here, modify the path of the img folder in data/config.py.
  • Then, generate a coco-style json.
python utils/pascal2coco.py --folder_path=/home/feiyu/Data/pascal_sbd
# Training.
python -m torch.distributed.launch --nproc_per_node=1 --master_port=$((RANDOM)) train.py --cfg=res50_pascal
# Evalution.
python eval.py --weight=weights/res50_pascal_120000.pth

Train custom datasets

  • Install labelme
pip install labelme
  • Use labelme to label your images, only ploygons are needed. The created json files are in the same folder with the images.
    Example 5
  • Prepare a 'labels.txt' like this, this first line: 'background' is always needed.
    Example 6
  • Prepare coco-style json, pass the paths of your image folder and the labels.txt. The 'custom_dataset' folder is a prepared example.
python utils/labelme2coco.py --img_dir=custom_dataset --label_name=cuatom_dataset/labels.txt
  • Edit CUSTOM_CLASSES in config.py.
    Example 7
    Note that if there's only one class, the CUSTOM_CLASSES should be like ('dog', ). The final comma is necessary to make it as a tuple, or the number of classes would be len('dog').
  • Choose a configuration ('res101_custom' or 'res50_custom') in config.py, modify the corresponding self.train_imgs and self.train_ann. If you need to validate, prepare the validation dataset by the same way.
  • Then train.
python -m torch.distributed.launch --nproc_per_node=1 --master_port=$((RANDOM)) train.py --cfg=res101_custom
  • Some parameters need to be taken care of by yourself:
  1. Training batch size, try not to use batch size smaller than 4.
  2. Anchor size, the anchor size should match with the object scale of your dataset.
  3. Total training steps, learning rate decay steps and the warm up step, these should be decided according to the dataset size, overwrite self.lr_steps, self.warmup_until in your configuration.

About

Minimal PyTorch implementation of YOLACT.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 98.6%
  • Cython 1.4%