Has anyone tried out video object detection models such as SELSA and MEGA. They seem to perform well on ImageNet Vid as shown by papers with code's leaderboard. Just curious!
", "messages": [ { "id": "1633711", "postDate": "12/31/2021 02:01:24", "content": "Has anyone tried out video object detection models such as SELSA and MEGA. They seem to perform well on ImageNet Vid as shown by papers with code's leaderboard. Just curious!
", "rawMarkdown": "Has anyone tried out video object detection models such as [SELSA](https://arxiv.org/abs/1907.06390) and [MEGA](https://openaccess.thecvf.com/content_CVPR_2020/papers/Chen_Memory_Enhanced_Global-Local_Aggregation_for_Video_Object_Detection_CVPR_2020_paper.pdf). They seem to perform well on ImageNet Vid as shown by [papers with code's leaderboard](https://paperswithcode.com/sota/video-object-detection-on-imagenet-vid). Just curious!", "votes": null }, { "id": "1634739", "postDate": "01/01/2022 05:26:50", "content": "They've definitely been on my radar on things to experiment with. I know SELSA is implemented in MMTracking and I'm currently working on the offline installation of MMtracking and a couple other OpenMMLab libraries for kaggle notebooks, you can find this as a dataset here. Another vid object detection that might be interesting is Temporal ROI Align
", "rawMarkdown": "They've definitely been on my radar on things to experiment with. I know SELSA is implemented in [MMTracking](https://github.com/open-mmlab/mmtracking) and I'm currently working on the offline installation of MMtracking and a couple other OpenMMLab libraries for kaggle notebooks, you can find this as a dataset [here](https://www.kaggle.com/maxvandijck/openmmlab-essential-repositories). Another vid object detection that might be interesting is [Temporal ROI Align](https://arxiv.org/abs/2109.03495)", "votes": null }, { "id": "1635431", "postDate": "01/01/2022 18:38:00", "content": "Nice. I was also thinking of combining REPP with yolo as that has been shown to boost performance significantly. Also, thanks for making an offline install of mmtracking.
", "rawMarkdown": "Nice. I was also thinking of combining [REPP](https://paperswithcode.com/paper/robust-and-efficient-post-processing-for) with yolo as that has been shown to boost performance significantly. Also, thanks for making an offline install of mmtracking.", "votes": null } ], "comments": [ { "id": 1634739, "author_name": "maxvandijck", "author_url": "", "post_date": "01/01/2022 05:26:50", "content": "They've definitely been on my radar on things to experiment with. I know SELSA is implemented in MMTracking and I'm currently working on the offline installation of MMtracking and a couple other OpenMMLab libraries for kaggle notebooks, you can find this as a dataset here. Another vid object detection that might be interesting is Temporal ROI Align
", "votes": null, "replies": [ { "id": 1635431, "author_name": "vexxingbanana", "author_url": "", "post_date": "01/01/2022 18:38:00", "content": "Nice. I was also thinking of combining REPP with yolo as that has been shown to boost performance significantly. Also, thanks for making an offline install of mmtracking.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1633711": "Has anyone tried out video object detection models such as [SELSA](https://arxiv.org/abs/1907.06390) and [MEGA](https://openaccess.thecvf.com/content_CVPR_2020/papers/Chen_Memory_Enhanced_Global-Local_Aggregation_for_Video_Object_Detection_CVPR_2020_paper.pdf). They seem to perform well on ImageNet Vid as shown by [papers with code's leaderboard](https://paperswithcode.com/sota/video-object-detection-on-imagenet-vid). Just curious!", "1634739": "They've definitely been on my radar on things to experiment with. I know SELSA is implemented in [MMTracking](https://github.com/open-mmlab/mmtracking) and I'm currently working on the offline installation of MMtracking and a couple other OpenMMLab libraries for kaggle notebooks, you can find this as a dataset [here](https://www.kaggle.com/maxvandijck/openmmlab-essential-repositories). Another vid object detection that might be interesting is [Temporal ROI Align](https://arxiv.org/abs/2109.03495)", "1635431": "Nice. I was also thinking of combining [REPP](https://paperswithcode.com/paper/robust-and-efficient-post-processing-for) with yolo as that has been shown to boost performance significantly. Also, thanks for making an offline install of mmtracking." }, "source": "meta" }