Datasets:
| # default is to download each image, square-crop, and then | |
| # resize to 512x512. Alter SIZE If desired | |
| # "objects" caption is tag style. Alternatively, use one | |
| # of the other files in the jsonl if you like. eg: | |
| # "llava38b", "internlm7b" | |
| # | |
| CAPTION=objects | |
| SIZE=512 | |
| # img2dataset is a python module | |
| img2dataset --url_list woman-croppable.jsonl.gz --input_format "jsonl.gz"\ | |
| --url_col "url" --caption_col "$CAPTION" --output_format files \ | |
| --output_folder pexels-woman-croppable \ | |
| --processes_count 4 --thread_count 4 \ | |
| --resize_mode center_crop --image_size=$SIZE \ | |
| --encode_quality=100 --compute_hash=md5 | |
| # compute hash is on to give an ability to cross reference | |