Skip to content

Latest commit

 

History

History
52 lines (43 loc) · 2.49 KB

README.md

File metadata and controls

52 lines (43 loc) · 2.49 KB
Raw

NTIRE 2023 Challenge on Efficient Super-Resolution @ CVPR 2023

How to test the baseline model?

  1. git clone https://github.com/ofsoundof/NTIRE2023_ESR.git
  2. Select the model you would like to test from run.sh 
    CUDA_VISIBLE_DEVICES=0 python test_demo.py --data_dir [path to your data dir] --save_dir [path to your save dir] --model_id 0
    • Be sure the change the directories --data_dir and --save_dir.

How to add your model to this baseline?

  1. Register your team in the Google Spreadsheet and get your team ID.
  2. Put your the code of your model in ./models/[Your_Team_ID]_[Your_Model_Name].py
    • Please add only one file in the folder ./models. Please do not add other submodules.
    • Please zero pad [Your_Team_ID] into two digits: e.g. 00, 01, 02
  3. Put the pretrained model in ./model_zoo/[Your_Team_ID]_[Your_Model_Name].[pth or pt or ckpt]
    • Please zero pad [Your_Team_ID] into two digits: e.g. 00, 01, 02
  4. Add your model to the model loader ./test_demo/select_model as follows: 
        elif model_id == [Your_Team_ID]:
        # define your model and load the checkpoint
    • Note: Please set the correct data_range, either 255.0 or 1.0
  5. Send us the command to download your code, e.g,
    • git clone [Your repository link]
    • We will do the following steps to add your code and model checkpoint to the repository.

How to calculate the number of parameters, FLOPs, and activations

    from utils.model_summary import get_model_flops, get_model_activation
    from models.team00_RFDN import RFDN
    model = RFDN()
    
    input_dim = (3, 256, 256)  # set the input dimension
    activations, num_conv = get_model_activation(model, input_dim)
    activations = activations / 10 ** 6
    print("{:>16s} : {:<.4f} [M]".format("#Activations", activations))
    print("{:>16s} : {:<d}".format("#Conv2d", num_conv))

    flops = get_model_flops(model, input_dim, False)
    flops = flops / 10 ** 9
    print("{:>16s} : {:<.4f} [G]".format("FLOPs", flops))

    num_parameters = sum(map(lambda x: x.numel(), model.parameters()))
    num_parameters = num_parameters / 10 ** 6
    print("{:>16s} : {:<.4f} [M]".format("#Params", num_parameters))

License and Acknowledgement

This code repository is release under MIT License.