Open KN1GHT9 opened 11 months ago
Hello. Here are the training temperature parameters for different datasets. MSCOCO images <---> MSCOCO captions: T=100 Flickr30k images <---> Flickr30k captions: T=125 MSCOCO images <---> Flickr30k captions: T=125 Flickr30k images <---> MSCOCO captions: T=125 MSCOCO images <---> SS captions: T=100 MSCOCO images <---> GCC captions: T=175
Please guide through the training process
after running python preprocess/coco/coco_train_images.py
(ta) D:\pp>python preprocess/coco/coco_train_images.py saving path:./data/coco/coco_ViT-L_14_train_images.pkl 4 captions loaded from json Traceback (most recent call last): File "preprocess/coco/coco_train_images.py", line 65, in <module> exit(main(args.clip_model_type, args.coco_path)) File "preprocess/coco/coco_train_images.py", line 30, in main img_ids.add(d["image_id"]) TypeError: string indices must be integers
this error what is the problem here
@Lihr747
and also the annotations json files are not clear how should I process Please help me
Congratulations on having your work accepted by ACM MM 2023! I have a question about the temperature, in your appendix the temperature is analyzed experimentally, but the final temperature parameter settings for each dataset are not given, is it possible to provide the training temperature parameters for different datasets as a reference?