Hi, I tested your code with both stable diffusion 1.5 and 2.1, and reconstructed results are so different. (Stable diffusion 2.1 seems doing much worse) Here are examples with DAVIS dataset samples.
car turn
(sd2.1)
(sd1.5)
bike
(sd2.1)
(sd1.5)
bear
(sd2.1)
(sd1.5)
Could you explain if there is any reason why the stable diffusion 2.1 performs pool?
Thanks for the your great work!
Hi, I tested your code with both stable diffusion 1.5 and 2.1, and reconstructed results are so different. (Stable diffusion 2.1 seems doing much worse) Here are examples with DAVIS dataset samples.
car turn (sd2.1)
(sd1.5)
![sample-500(1 5)](https://user-images.githubusercontent.com/107248364/235876595-3eff4b6f-c96e-4dd1-97a3-940b188c725d.gif)
bike (sd2.1)
(sd1.5)
![sample-100_1_5_](https://user-images.githubusercontent.com/107248364/235876935-f4847e00-7521-4c56-bc9b-6b42058c14b8.gif)
bear (sd2.1)
(sd1.5)
![sample-100_1_5](https://user-images.githubusercontent.com/107248364/235877268-b2585196-5851-4c35-8ee3-6eaade9cb80f.gif)
Could you explain if there is any reason why the stable diffusion 2.1 performs pool? Thanks for the your great work!