Datasets:
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,27 +1,57 @@
|
|
| 1 |
---
|
| 2 |
-
license:
|
| 3 |
-
pretty_name:
|
| 4 |
tags:
|
| 5 |
- synthetic
|
| 6 |
- human-pose-shape
|
|
|
|
|
|
|
| 7 |
- computer-vision
|
| 8 |
---
|
| 9 |
|
| 10 |
-
#
|
| 11 |
|
| 12 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
|
| 14 |
## Structure
|
| 15 |
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
|
| 20 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
|
| 22 |
-
For a dataset chunk named `human_dataset_v35_1`:
|
| 23 |
-
- images are in `shards/human_dataset_v35_1/`
|
| 24 |
-
- labels are in `labels/human_dataset_v35_1_combined.npz`
|
| 25 |
|
| 26 |
## Citation
|
| 27 |
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
license: bsd-3-clause
|
| 3 |
+
pretty_name: Infinigen Humans Dataset
|
| 4 |
tags:
|
| 5 |
- synthetic
|
| 6 |
- human-pose-shape
|
| 7 |
+
- 3d-human-pose
|
| 8 |
+
- smplx
|
| 9 |
- computer-vision
|
| 10 |
---
|
| 11 |
|
| 12 |
+
# Infinigen Humans Dataset
|
| 13 |
|
| 14 |
+
Synthetic human image dataset generated for:
|
| 15 |
+
|
| 16 |
+
**On the Role of Visual Realism in Synthetic Data for Estimating 3D Human Pose and Shape**
|
| 17 |
+
|
| 18 |
+
Paper: TODO
|
| 19 |
+
Code: https://github.com/princeton-vl/infinigen-human
|
| 20 |
|
| 21 |
## Structure
|
| 22 |
|
| 23 |
+
```text
|
| 24 |
+
shards/
|
| 25 |
+
<dataset_name>/
|
| 26 |
+
shard-*.tar
|
| 27 |
+
|
| 28 |
+
labels/
|
| 29 |
+
<dataset_name>_combined.npz
|
| 30 |
+
|
| 31 |
+
manifest.csv
|
| 32 |
+
```
|
| 33 |
+
|
| 34 |
+
`manifest.csv` maps each shard directory to its matching label file.
|
| 35 |
+
|
| 36 |
+
## Labels
|
| 37 |
+
|
| 38 |
+
Each `.npz` label file contains frame level annotations, including:
|
| 39 |
|
| 40 |
+
- `imgname`: image path/name matching entries inside the tar shards
|
| 41 |
+
- `center`: 2D bounding-box center
|
| 42 |
+
- `pose_cam`: SMPL-X pose in camera coordinates
|
| 43 |
+
- `pose_world`: SMPL-X pose in world coordinates
|
| 44 |
+
- `shape`: SMPL-X shape coefficients
|
| 45 |
+
- `trans_cam`: camera-space translation
|
| 46 |
+
- `trans_world`: world-space translation
|
| 47 |
+
- `gtkps`: 2D keypoints with visibility
|
| 48 |
+
- `cam_int`: camera intrinsics
|
| 49 |
+
- `cam_ext`: camera extrinsics
|
| 50 |
+
- `proj_verts`: projected downsampled SMPL-X vertices
|
| 51 |
|
|
|
|
|
|
|
|
|
|
| 52 |
|
| 53 |
## Citation
|
| 54 |
|
| 55 |
+
```bibtex
|
| 56 |
+
TODO
|
| 57 |
+
```
|