Shilin
Thanks for your interest. Let's consider two matrices: Matrix A: | a b | | a b | Matrix B: | c d | | d c | Multiplying A...
Yes, you can do in that way. I believe their performance would be similar, but your way would be much more efficient than mine. I didn’t come up with this...
Yes, you can do in that way. I believe their performance would be similar, but your way would be much more efficient than mine. I didn’t come up with this...
Thanks for your help!
Hi, thanks for your interest. Yes, the dataset used to evaluate the FID is MSCOCO. Indeed, this may not be an optimal choice. We aim to see if the generation...
Hi, the hyperparameters for celebrity 1,5,10,100 are included in `configs/celebrity`, and you also can find them in Table 10, `preserve_weight` (in config) equals `lambda_1 * lambda_3` (in Table 10).
Hi, thanks for your interest. The clip_score evaluation script is included in `metrics/evaluate_clip_score.py`, and you could input the prompt list `prompts_csv/art_100_concepts.csv` to evaluate the erasure of art style. We apply...
Yes, you are right. Just small modifications work well.
You may check the instructions from [Grounded-SAM](https://github.com/IDEA-Research/Grounded-Segment-Anything.git) for more details
Hi, to fit your case, probably you can try to reduce the `max_training_step` from 120 to 50-60, and increase both the `train_preserve_scale` and `fuse_preserve_scale` to 1e-4 or 1e-5.