1: Train with the existing models and satndard datastes¶

To evaluate a model’s accuracy, one usually tests the model on some standard datasets. FGVCLib supports the public datasets including CUB_200_2011. This section will show how to test existing models on supported datasets.

The basic steps are as below: 1.Prepare the dataset 2.Prepare a config 3.Train, test models on the dataset

The existing models¶

We provide a variety of existing methods, they are baseline_resnet50, MCL, PMG, PMG_v2, API-Net, CAL, PIM, TransFG.

In the future we will continue to reproduce new methods and add them into FGVCLib.

Prepare the dataset¶

We provide the CUB-200-2011, and we split the dataset into train and test folder.

e.g., CUB-200-2011 dataset

  -/birds/train
	         └─── 001.Black_footed_Albatross
	                   └─── Black_Footed_Albatross_0001_796111.jpg
	                   └─── ...
	         └─── 002.Laysan_Albatross
	         └─── 003.Sooty_Albatross
	         └─── ...
   -/birds/test	
             └─── ...         

If you have prepared the dataset, you can skip the step1.

step1: open the “/fgvclib/datasets/cub.py”, and modify the class CUB_200_2011: __init__ : download:bool=False to class CUB_200_2011: __init__ : download:bool=True

The parameter 'download' controls whether the dataset is downloaded. Directly downloading CUB dataset by setting download=True. Default is False.

step2: open the “/configs/xxx/xxx.yml”, and replace the DATASET-ROOT with your own path.

Train¶

step1: open the “/configs/xxx/xxx.yml”, and replace the WEIGHT-SAVE_DIR with your own path. step2: open the “/configs/xxx/xxx.yml”, and check the configs about the model. You can change the configs by yourself. stpe3: execute main program to train.

python main.py --config configs/resnet/resnet50.yml

There are several arguments to control the program.

‘–config’: the path of configuration file.
‘–task’: train or predict. The default is train.
‘–device’: two choices are cuda and cpu. The default is cuda.
‘–world-size’: the number of distributed processes. The default is 4.
‘–dist-url’: url used to set up distributed training. The default is ‘env://’.

If you want to run it on cpu, you should execute the following：

python main.py --config configs/resnet/resnet50.yml --device cpu

Test¶

python main.py --config configs/resnet/resnet50.yml --task predict