yolact-test

A test version of Yolact in PyTorch for instance segmentation

Acknowledgement

• This repository references dbolya's work.

Dataset

• Check COCO 2017 dataset

  python dataset_player.py
  python dataset_player.py --training
  

• Check KITTI dataset

  python dataset_player.py --dataset=kitti_dataset
  python dataset_player.py --dataset=kitti_dataset --training
  

• Check SEUMM HQ LWIR dataset

  python dataset_player.py --dataset=seumm_hq_lwir_dataset
  python dataset_player.py --dataset=seumm_hq_lwir_dataset --training
  

• Check SEUMM LWIR dataset

  python dataset_player.py --dataset=seumm_lwir_dataset
  python dataset_player.py --dataset=seumm_lwir_dataset --training
  

• Check LAJI 4692 dataset

  python dataset_player.py --dataset=laji_4692_dataset
  python dataset_player.py --dataset=laji_4692_dataset --training
  

Training

• Train on COCO 2017 dataset

  python train.py --config=yolact_resnet50_config
  

• Train on KITTI dataset

  python train.py --config=yolact_resnet50_config --dataset=kitti_dataset
  

• Train on SEUMM HQ LWIR dataset

  python train.py --config=yolact_resnet50_config --dataset=seumm_hq_lwir_dataset
  

• Train on SEUMM LWIR dataset

  python train.py --config=yolact_resnet50_config --dataset=seumm_lwir_dataset
  

• Train on LAJI 4692 dataset

  python train.py --config=yolact_resnet50_config --dataset=laji_4692_dataset --batch_size=4 --lr=0.0005
  

Evaluation

• Evaluate on COCO 2017 dataset

  python eval.py --trained_model=weights/coco/yolact_resnet50_25_380000.pth
  python eval.py --trained_model=weights/coco/yolact_resnet50_25_380000.pth --display
  

• Evaluate on KITTI dataset

  python eval.py --dataset=kitti_dataset --trained_model=weights/kitti/yolact_resnet50_107_60000.pth
  python eval.py --dataset=kitti_dataset --trained_model=weights/kitti/yolact_resnet50_107_60000.pth --display
  

• Evaluate on SEUMM HQ LWIR dataset

  python eval.py --dataset=seumm_hq_lwir_dataset --trained_model=weights/seumm_hq_lwir/yolact_resnet50_256_60000.pth
  python eval.py --dataset=seumm_hq_lwir_dataset --trained_model=weights/seumm_hq_lwir/yolact_resnet50_256_60000.pth --display
  

• Evaluate on SEUMM LWIR dataset

  python eval.py --dataset=seumm_lwir_dataset --trained_model=weights/seumm_lwir/yolact_resnet50_72_60000.pth
  python eval.py --dataset=seumm_lwir_dataset --trained_model=weights/seumm_lwir/yolact_resnet50_72_60000.pth --display
  

• Evaluate on LAJI 4692 dataset

  python eval.py --dataset=laji_4692_dataset --trained_model=weights/laji_4692/yolact_resnet50_205_160000.pth
  python eval.py --dataset=laji_4692_dataset --trained_model=weights/laji_4692/yolact_resnet50_205_160000.pth --display --top_k=50
  

• The result should be

Backbone	Dataset	Iter	val [email protected]	val [email protected]:.95B	val [email protected]	val [email protected]:.95M
ResNet50	COCO	380k	46.56	27.35	42.75	25.78
ResNet50	KITTI	60k	44.67	24.23	39.55	22.34
ResNet50	SEUMM-HQ-L	60k	86.66	49.05	78.74	42.26
ResNet50	SEUMM-L	60k	72.67	40.76	64.98	37.37
ResNet50	LAJI-4692	80k	46.08	21.78	36.21	16.04

Demo

• Run a demo with COCO 2017 model

  python eval.py --trained_model=weights/coco/yolact_resnet50_25_380000.pth --image=my_image.jpeg --score_threshold=0.25 --top_k=20
  python eval.py --trained_model=weights/coco/yolact_resnet50_25_380000.pth --images=test_images:outputs --score_threshold=0.25 --top_k=20
  

• Run a demo with KITTI model

  python eval.py --dataset=kitti_dataset --trained_model=weights/kitti/yolact_resnet50_107_60000.pth --image=my_image.jpeg --score_threshold=0.25 --top_k=20
  python eval.py --dataset=kitti_dataset --trained_model=weights/kitti/yolact_resnet50_107_60000.pth --images=test_images:outputs --score_threshold=0.25 --top_k=20
  

• Run a demo with SEUMM HQ LWIR model

  python eval.py --dataset=seumm_hq_lwir_dataset --trained_model=weights/seumm_hq_lwir/yolact_resnet50_256_60000.pth --image=my_image.jpeg --score_threshold=0.25 --top_k=20
  python eval.py --dataset=seumm_hq_lwir_dataset --trained_model=weights/seumm_hq_lwir/yolact_resnet50_256_60000.pth --images=test_images:outputs --score_threshold=0.25 --top_k=20
  

• Run a demo with SEUMM LWIR model

  python eval.py --dataset=seumm_lwir_dataset --trained_model=weights/seumm_lwir/yolact_resnet50_72_60000.pth --image=my_image.jpeg --score_threshold=0.25 --top_k=20
  python eval.py --dataset=seumm_lwir_dataset --trained_model=weights/seumm_lwir/yolact_resnet50_72_60000.pth --images=test_images:outputs --score_threshold=0.25 --top_k=20
  

• Run a demo with LAJI 4692 model

  python eval.py --dataset=laji_4692_dataset --trained_model=weights/laji_4692/yolact_resnet50_205_160000.pth --image=my_image.jpeg --score_threshold=0.25 --top_k=50
  python eval.py --dataset=laji_4692_dataset --trained_model=weights/laji_4692/yolact_resnet50_205_160000.pth --images=test_images:outputs --score_threshold=0.25 --top_k=50