search

seoungwugoh / STM

Video Object Segmentation using Space-Time Memory Networks
405 stars 81 forks source link

questions about Ms and Fs #21

Open cernykisss opened 4 years ago

cernykisss commented 4 years ago

Hallo! Thanks for your codes! I appreciate you very much! After i read your codes carefuully,i've noticed that there are two tensors called Es and Ms.

Es = torch.zeros_like(Ms) Es[:,:,0] = Ms[:,:,0]

Could you tell me their accurate meaning?

wish you a good day

sjb961121 commented 4 years ago

Hallo! Thanks for your codes! I appreciate you very much! After i read your codes carefuully,i've noticed that there are two tensors called Es and Ms.

Es = torch.zeros_like(Ms) Es[:,:,0] = Ms[:,:,0]

Could you tell me their accurate meaning?

wish you a good day

I have the same question,Have you solved it? Could you help me?💋

seoungwugoh commented 4 years ago

Hi @cernykisss @sjb961121 Originally, Ms contains the ground-truth mask and Es contains estimations. Ms was used for training but it no longer effective during inference. So, in the current demo code, you can consider Ms is just legacy doing nothing.

cernykisss commented 4 years ago

Thanks for your reply！ I've read your paper and knew that there are a Siamese Encoder consisting of two identical encoders in the model.
I'm still wondering the meaning of Encoder_q and Encoder_m in your model.py.

seoungwugoh commented 4 years ago

In the paper (STM, http://openaccess.thecvf.com/content_ICCV_2019/papers/Oh_Video_Object_Segmentation_Using_Space-Time_Memory_Networks_ICCV_2019_paper.pdf), no Siamese encoder is used. There are two different encoder each for query and memory image, respectively. You may confused with my previous work (RGMP, http://openaccess.thecvf.com/content_cvpr_2018/papers/Oh_Fast_Video_Object_CVPR_2018_paper.pdf)

cernykisss commented 4 years ago

답장 해주셔서 매우 감사합니다！ 나는 신문을 잘못 읽었다.

cernykisss commented 4 years ago

Hello! Sorry for interrupting you again. I've read your paper about STM. But the I want to know more about multi object segment. Your paper says 'More details are included in the supplementary materials'. Where is the supplementary materials? Could you give me a link? Waiting for your answer!

seoungwugoh commented 4 years ago

@cernykisss You can download it through the download link in the GitHub page. https://openaccess.thecvf.com/content_ICCV_2019/supplemental/Oh_Video_Object_Segmentation_ICCV_2019_supplemental.zip

cernykisss commented 4 years ago

thank you！