Assembly101

Assembly101 is a procedural activity dataset featuring 4321 videos of people assembling and disassembling 101 "take-apart" toy vehicles. Participants work without fixed instructions, and the sequences feature rich and natural variations in action ordering, mistakes, and corrections. Assembly101 is the first multi-view action dataset, with simultaneous static (8) and egocentric (4) recordings. Sequences are annotated with more than 100K coarse and 1M fine-grained action segments, and 18M 3D hand poses. We benchmark on three action understanding tasks: recognition, anticipation and temporal segmentation. Additionally, we propose a novel task of detecting mistakes. The unique recording format and rich set of annotations allow us to investigate generalization to new toys, cross-view transfer, long-tailed distributions, and pose vs. appearance. We envision that Assembly101 will serve as a new challenge to investigate various activity understanding problems.

The repo structure is as follows:

• recordings: containing all 362 recordings (12 views per recording - 8 fixed and 4 egocentric)
• AssemblyPoses.zip: containing 2D, 3D hand poses at 60fps, camera extrinsics and positions for all the 12 views, and hand-object crops (defined as 125% of the rectangular area enclosed by the 2D hand pose)
• poses@60fps: containing 3D hand poses generated at 60fps (please use 2D/3D hand poses in AssemblyPoses.zip for the latest cleaned version)
• TSM_features: containing per-frame visual features extracted by TSM
• DINOv2_features.zip: containing per-frame DINOv2 (ViT-g/14) features for both full frame and the hand-object crops (refer to HandFormer for more details)

Folder structure of the Assembly101 recordings:

recordings
├── nusar-2021_action_both_9086-c14a_9086_user_id_2021-02-16_153910
│   ├── C10095_rgb.mp4
│   ├── C10115_rgb.mp4
│   ├── C10118_rgb.mp4
│   ├── C10119_rgb.mp4
│   ├── C10379_rgb.mp4
│   ├── C10390_rgb.mp4
│   ├── C10395_rgb.mp4
│   ├── C10404_rgb.mp4
│   ├──HMC_21176875_mono10bit.mp4
│   ├──HMC_21176623_mono10bit.mp4
│   ├──HMC_21110305_mono10bit.mp4
│   ├──HMC_21179183_mono10bit.mp4
├── nusar-2021_action_both_9071-b06b_9071_user_id_2021-02-11_100739
│   ├── C10095_rgb.mp4
│   ├── .
│   ├── .
│   ├── .
.
.
.

Which camera_IDs correspond to which views ?

C10095_rgb : v1
C10115_rgb : v2
C10118_rgb : v3
C10119_rgb : v4
C10379_rgb : v5
C10390_rgb : v6
C10395_rgb : v7
C10404_rgb : v8

HMC_84346135_mono10bit or HMC_21176875_mono10bit : e1
HMC_84347414_mono10bit or HMC_21176623_mono10bit : e2
HMC_84355350_mono10bit or HMC_21110305_mono10bit : e3
HMC_84358933_mono10bit or HMC_21179183_mono10bit : e4

Egocentric views camera ID mapping:

For some videos the camera ID for egocentric (monochrome) views starts with HMC_8... and for some they start with HMC_2.... This occurs due to the usage of different cameras for some of the videos. The camera ID mapping is as follows:

HMC_84346135_mono10bit : HMC_21176875_mono10bit
HMC_84347414_mono10bit : HMC_21176623_mono10bit
HMC_84355350_mono10bit : HMC_21110305_mono10bit
HMC_84358933_mono10bit : HMC_21179183_mono10bit

License

Assembly101 is licensed by us under the Creative Commons Attribution-NonCommerial 4.0 International License, found here. The terms are :

• Attribution : You must give appropriate credit, provide a link to the license, and indicate if changes were made. You may do so in any reasonable manner, but not in any way that suggests the licensor endorses you or your use.
• NonCommercial : You may not use the material for commercial purposes.