NUS Computer Vision and Machine Learning Lab
@cvml_nus
Followers
173
Following
6
Media
8
Statuses
11
Computer Vision and Machine Learning Group at the National University of Singapore
Singapore
Joined March 2022
HandFormer sets new SoTA in action recognition for Assembly101 and H2O. Our pose-only variant beats existing skeleton-based methods and, when supplemented with even one RGB frame, can outperform SoTA RGB models on Assembly101.
0
0
1
RGB models are great, but they can be slow and costly. 3D hand poses offer a faster and compact solution, but they miss out on object information. HandFormer combines both and gives us the best of both worlds - higher temporal resolution and scene context.
1
0
1
Check out our latest #ECCV2024 work HandFormer where we delve deeper into 3D hand poses to recognize hand-object interactions for always-on AR/VR applications!. Team: @m_s_shamil @dibschat_ @famesener Shugao Ma @angelayao101 . 🔗 Learn more at
1
4
3
Assembly101 has also opened the door for large-scale hand-object interaction understanding in 3D with followup works like AssemblyHands ( from @tkhkaeio and HandFormer ( from @m_s_shamil.
0
1
2
Our proposed mistake detection task has been well recognized in the community and followup works [Ding et al.] and [@AlessandroFlabo @anfurnari @GMFarinella] delve deeper into this.
1
1
3
Assembly101 was recognized at the #EGOVIS 2022/23 distinguished paper awards! Congrats to the whole team! . @famesener @dibschat_ @angelayao101
#CVPR2024 @CVPR
1
1
7
Check out our new #CVPR2022 paper that generates temporally coherent UV coordinates for loose clothing. Our method is not constrained by human body outlines. It can be used to flexibly synthesize new looks and modified visual styles. Paper:
1
2
4
RT @angelayao101: Our #CVPR22 paper presents Assembly101, a new large-scale procedural activity dataset with 513 hours of video of people a….
0
2
0
Check out our new #CVPR2022 paper on semi-supervised video object segmentation. We propose an efficient plug-and-play acceleration framework by exploiting the temporal redundancies in the video presented by the compressed bitstream. Project:
0
2
3
