Training-Code-of-STM
Training-Code-of-STM copied to clipboard
Training code of Spatial Time Memory Network. Semi-supervised video object segmentation.
The weight link of Resnet10 is invalid, you can release the weight of Resnet101 again
Hi, can this training code also reproduce the performance on YoutubeVOS? Is there any evaluation code for that also?
Thanks for your good work! I have the following questions: 1, why is the Davis dataset used when the model is pre-trained on the Coco dataset, and 2, line 90...
Thanks for your effort and for sharing the code. Is it possible to fine-tune the pre-trained models to a custom dataset? How? And Does the fine-tuning make the performance better...
could you please explain the mean of in_m & in_o in forward function of Encoder_M?
Could you please add some annotations for function? it is difficult to read and understand.
which YouTube-VOS dataset do you use to train the network? 2018 or 2019?
thanks for your sharing. but how can i implement multi-gpu for training?and how to raise batchsize to more than 1? thanks !
as the title
Hi, I have a question. How many iterations were needed to obtain J&F = 69.5 in training stage 1 with only MS-COCO? Thanks!