Closed Tokyo-dream closed 9 months ago
Thanks for this amazing work !
Hi @Tokyo-dream. Thanks for your interest in our work!
As you’ve noted, the training for each object is essentially an optimization process, which can be easily affected by the randomness. I was also encountering the unstable training issue you mentioned in my practice. To get better results, it sometimes helps to tweak the learning rate a little bit or just train multiple times.
From my experience, there are varying degrees of training difficulty across these ten cases. Objects with relatively thin parts present a particular challenge as there are fewer opportunities for these regions to establish the correct correspondence. This can be considered a limitation inherent to our method. Increasing the number of views can, in many cases, help to enhance the likelihood of obtaining consistent positive results.
This is the result trained through pretrain/parsed.yaml This is the result trained by pretrained model How can I set the parameters to make the mobile part more accurate?