Open NEO-237 opened 3 weeks ago
Thanks for your interest!
CLIP-ES is based on the pre-trained CLIP model, which is a powerful foundation model trained on large-scale data and can apply to most common categories or scenarios. CLIP-ES is training-free and if your dataset is unique and uncommon, you may try the following ways: 1) carefully custome the text description for each category you used; 2) fine-tune the original CLIP model using your own data; 3) try other training-based WSSS methods.
Thanks a lot!
I used CLIP-ES on my own dataset, but it didn't work very well for generatig CAMs. Is there any way to train this model to better fit my dataset? Thanks.