If you want to train the model, you should provide the dataset path, in the dataset path, a separate gt text file should be provided for each image and run python multigpu_train.py --gpu_list=0 ...