Open qizhongjian opened 4 years ago
I use the batch_size = 3 and only a gpu.
Hello, there's the same question for me now. The pre-trained model on CUB shows 4.65 on IS. So do you solve the problem now?
@shangshng no idea,now.
Hi, @shangshng @qizhongjian
Could you please show your configurations ( batch size, branch number, and pytorch version). I generate about 30,000 images using batch size = 3 and branch number = 3. I get 4.69 std: 0.07 on CUB dataset.
Besides, have your tried batch size = 10?
Hi, @shangshng @qizhongjian
Could you please show your configurations ( batch size, branch number, and pytorch version). I generate about 30,000 images using batch size = 3 and branch number = 3. I get 4.69 std: 0.07 on CUB dataset.
Besides, have your tried batch size = 10?
Hello, I'm using pip install torch==0.4.2 Will it affect the result?
I use CUDA Version:10.1 ,and there is no error when running code, but will it affect the result? I get IS 4.63 on CUB,too...
I use CUDA Version:10.1 ,and there is no error when running code, but will it affect the result? I get IS 4.63 on CUB,too...
How about other evaluation metrics (R and FID) ?
I use CUDA Version:10.1 ,and there is no error when running code, but will it affect the result? I get IS 4.63 on CUB,too...
How about other evaluation metrics (R and FID) ?
Thanks for your reply. I get 20 in FID, and I don't know how to calculate the R precision, Can you help me about that... Thanks again!
I use CUDA Version:10.1 ,and there is no error when running code, but will it affect the result? I get IS 4.63 on CUB,too...
How about other evaluation metrics (R and FID) ?
Hello sir I train the DM Model in my computer.. and When epoch approaches 150,the images generated by G are very similar, is that model collapse?should I train more or do some change in parameter? (batch size:10 branch num: 3 D&G LR: 0.0002) Thanks
I use CUDA Version:10.1 ,and there is no error when running code, but will it affect the result? I get IS 4.63 on CUB,too...
How about other evaluation metrics (R and FID) ?
Thanks for your reply. I get 20 in FID, and I don't know how to calculate the R precision, Can you help me about that... Thanks again!
The source code will calculate the R precision when you evaluate the model.
python main.py --cfg cfg/eval_bird.yml --gpu 0
Currently, we do not support evaluate R precision during training.
I use CUDA Version:10.1 ,and there is no error when running code, but will it affect the result? I get IS 4.63 on CUB,too...
How about other evaluation metrics (R and FID) ?
Hello sir I train the DM Model in my computer.. and When epoch approaches 150,the images generated by G are very similar, is that model collapse?should I train more or do some change in parameter? (batch size:10 branch num: 3 D&G LR: 0.0002) Thanks
I have no idea have to address this issue. Is there anything you modified? Maybe you can try running this code on other hardware...
I have no idea have to address this issue. Is there anything you modified? Maybe you can try running this code on other hardware...
Thanks for your suggestion. I found that the graphics card will affect the results. Which version of GPU you are using (I used 1080ti) Thanks again!
I have evaluate the pre-trained model provided by author, but i only got IS 4.62 on CUB. I want know is there any tricks in code?