search

google-research / pix2seq

Pix2Seq codebase: multi-tasks with generative modeling (autoregressive and diffusion)
Apache License 2.0
857 stars 71 forks source link

How much time needed in the VOS task? #37

Open isksjsksk opened 1 year ago

isksjsksk commented 1 year ago

DDPM is known to be time-consuming and I am not sure if it is suitable for video segmentation tasks. So I wonder how much time is needed to for video object segmentation

chentingpc commented 1 year ago

overall time = encoder single pass time + decoder single pass time * iterations where iterations can be 5 or 50 with varying performances (see table 1 or figure 7 of https://arxiv.org/pdf/2210.06366.pdf).

isksjsksk commented 1 year ago

Hi, I start to run pix2seq-Drecently and have encountered some problems.

Firstly, when attempting to train on the COCO 2017 panoptic dataset, I noticed that the files in the data and metrics folders do not seem to contain the complete content for the coco image panoptic segmentation. Additionally, while trying on videos, I encountered another error: "dataset_config = copy.deepcopy(dataset_configs.dataset_configs[ds_name]), KeyError: 'kittistep_vps'." Upon investigating the dataset_configs.py file, I discovered that it also does not include the configuration for panoptic segmentation.

Thanks.

 

------------------ 原始邮件 ------------------ 发件人: "google-research/pix2seq" @.>; 发送时间: 2023年4月29日(星期六) 上午7:21 @.>; @.**@.>; 主题: Re: [google-research/pix2seq] How much time needed in the VOS task? (Issue #37)

overall time = encoder single pass time + decoder single pass time * iterations where iterations can be 5 or 50 with varying performances (see table 1 or figure 7 of https://arxiv.org/pdf/2210.06366.pdf).

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>