| name | pretrain | resolution | acc | #params | links-model | links-ema |
|---|---|---|---|---|---|---|
| model_stnvit_mixs | 384 | x | https://drive.google.com/uc?export=download&id=16RLc7MQmCXuRn_PuN31Js5OsgRIECRBH | https://drive.google.com/uc?export=download&id=10rmlJGUQ5LGkRppXQj1qevFmFtiuk3EU | ||
| model_swin | 384 | x | https://drive.google.com/uc?export=download&id=1R5SSLjuUGA7NdCNVkzMinO59Vo9U38lh | https://drive.google.com/uc?export=download&id=1tqchfh4MpLnI57ZaqJDQKmG2-qT9HcP5 | ||
| model_swin_mixs_tmax101 | 384 | x | https://drive.google.com/uc?export=download&id=1yhbiOHs4KH4frMLd3vdYADpjvNemJ378 | https://drive.google.com/uc?export=download&id=1_h60KHU3yFRZjectA2GAP7r7tV-znSSg | ||
| model_convnext_384_tmax101 | 384 | x | https://drive.google.com/uc?export=download&id=1mzRVoejVL_pnqjN28XcKLNnqqnTNVdUz | https://drive.google.com/uc?export=download&id=1II7Vd6CZqyXNywSSICXFmMdYwQAAIkkZ |
If you want to used the pretrained model to evaluate the result directly, please save the model.pth, ema.pth into ./saved/<model_name>/model.pth, which is the default path for evaluate.py to get model preatrined werights.
git clone https://github.com/jimmylin0979/AICUP2022-OrchidClassifier.git
cd AICUP2022-OrchidClassifier
pip install -r requirements.txtConfiguration files of models used for experiments are located inside ./config.py file. You may edit these files depending upon the location of datasets, ratio of how to split train/valid set, .., and so on.
Below is the training / testing steps for the one who wants to train the model via command line.
There is also a ipynb file, so you can simply run the ipynb file easily. (But i highly recommanded to train via command line)
The param <model_name> is the folder name that will be created to store the whole model information.
CUDA_VISIBLE_DEVICES=0 python main.py --logdir <model_name>Noted that default location of datasets in the config.py are ./data/dataset/train.
Please make change accordingly.
Like the param in training, the param <model_name> is the folder name that you want to use to evaluate the dataset.
CUDA_VISIBLE_DEVICES=0 python evaluate.py --model <model_name>Ensemble the prediction results of all models.
python ensemble_easy.py