Closed yinglang closed 6 days ago
Thank you for interesting about our work. I added the processed dataset samples on dataset After unzip dataset, you change config.train_img_path, config.val_img_path, config.train_json, config.val_json at src/common/configs_stage2_dino_codellama.py to your unzipped path.
Thanks very much!That's much helpful for me. By the way, is this processing done by directly taking the bounding boxes of all "text", "underlay", "logo" and "embellishment" as the mask input for LAMA?
Yes. we made using off-the-shelf model LAMA Injecting ground truth bounding box of Cgl-dataset.
Much Thanks!!! i got it.
That's a very nice work! I want to know if the clean image after LAMA processing will be opened, or the processing script will be opened, I need to train on my dataset, so the processing script is essential for me. It will be much helpful if the script is available. Much looking forward for your response.