Views
No views yet
pip install -r requirements.txt 1PAN
2└── dataset
3 └── benchmark
4 ├── Set5
5 ├── HR
6 | ├── baby.png
7 | ├── ...
8 └── LR_bicubic
9 └──X2
10 ├──babyx2.png
11 ├── ...
12 ├── Set14
13 ├── ... infer_onnx.py on how to use1 parser = argparse.ArgumentParser(description='PAN SR')
2 parser.add_argument('--onnx_path',
3 type=str,
4 default='PAN_int8.onnx',
5 help='Onnx path')
6 parser.add_argument('--image_path',
7 type=str,
8 default='test_data/test.png',
9 help='Path to your input image.')
10 parser.add_argument('--output_path',
11 type=str,
12 default='test_data/sr.png',
13 help='Path to your output image.')
14 parser.add_argument('--provider_config',
15 type=str,
16 default="vaip_config.json",
17 help="Path of the config file for seting provider_options.")
18 parser.add_argument('--ipu', action='store_true', help='Use Ipu for interence.')
19
20 args = parser.parse_args()
21
22 onnx_file_name = args.onnx_path
23 image_path = args.image_path
24 output_path = args.output_path
25
26 if args.ipu:
27 providers = ["VitisAIExecutionProvider"]
28 provider_options = [{"config_file": args.provider_config}]
29 else:
30 providers = ['CPUExecutionProvider']
31 provider_options = None
32 ort_session = onnxruntime.InferenceSession(onnx_file_name, providers=providers, provider_options=provider_options)
33
34 lr = cv2.imread(image_path)[np.newaxis,:,:,:].transpose((0,3,1,2)).astype(np.float32)
35 sr = tiling_inference(ort_session, lr, 8, (56, 56))
36 sr = np.clip(sr, 0, 255)
37 sr = sr.squeeze().transpose((1,2,0)).astype(np.uint8)
38 sr = cv2.imwrite(output_path, sr)python infer_onnx.py --onnx_path PAN_int8.onnx --image_path /Path/To/Your/Image --ipu --provider_config Path\To\vaip_config.jsonpython eval_onnx.py --onnx_path PAN_int8.onnx --data_test Set5 --ipu --provider_config Path\To\vaip_config.json| Method | Scale | Flops | Set5 |
|---|---|---|---|
| PAN (float) | X2 | 141G | 38.00 / 0.961 |
| PAN_amd (float) | X2 | 141G | 37.859 / 0.960 |
| PAN_amd (int8) | X2 | 141G | 37.18 / 0.952 |
1@inproceedings{zhao2020efficient,
2 title={Efficient image super-resolution using pixel attention},
3 author={Zhao, Hengyuan and Kong, Xiangtao and He, Jingwen and Qiao, Yu and Dong, Chao},
4 booktitle={European Conference on Computer Vision},
5 pages={56--72},
6 year={2020},
7 organization={Springer}
8}