How to train on the custom data set?

[henghuiding]

How to train on the custom data set?

[ntuLC]

Hi Yunlong,

Briefly speaking, you need to prepare a dataset annotation JSON file, then specify the path to the image folder and target mask folder in the config file. For details please check:

Vision-Language-Transformer/data/data_process_v2.py

Line 77 in 9b24015

def prepare_dataset(dataset, splits, output_dir, generate_mask=False):

Vision-Language-Transformer/loader/loader.py

Line 93 in 9b24015

def get_random_data(ref, input_shape, embed, config, train_mode=True, max_boxes=1):

[YunLongPan]

Thank you! At 2021-10-06 18:30:04, "ntuLC" ***@***.***> wrote: Hi Yunlong, Briefly speaking, you need to prepare a dataset annotation JSON file, then specify the path to the image folder and target mask folder in the config file. For details please check: https://github.com/henghuiding/Vision-Language-Transformer/blob/9b24015566fa820e3eddbbd8942fa44512ec1b3c/data/data_process_v2.py#L77 https://github.com/henghuiding/Vision-Language-Transformer/blob/9b24015566fa820e3eddbbd8942fa44512ec1b3c/loader/loader.py#L93 — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.