Thank you for the code. It works seamlessly, and it is easy to understand. However, I cannot run the training script and get a good decomposition. However, the ckpt you have provided is working for me.
Can you please help me train the model so that I can do good decomposition on the CLEVR dataset? Is there some particular seed value for which it works?
Hi Goirik, in our experiments, a run with batch size 32 and 10000 steps could achieve good decomposition quality for CLEVR. How long did you train the model for?
Dear authors,
Thank you for the code. It works seamlessly, and it is easy to understand. However, I cannot run the training script and get a good decomposition. However, the ckpt you have provided is working for me.
Can you please help me train the model so that I can do good decomposition on the CLEVR dataset? Is there some particular seed value for which it works?
Regards,