Commit
Β·
2d3338b
1
Parent(s):
dd6884f
add lightspeed
Browse files- README.md +56 -3
- lightspeed.zip +3 -0
README.md
CHANGED
@@ -15,6 +15,10 @@ task_categories:
|
|
15 |
[Ming-Yu Liu<sup>1</sup>](https://mingyuliu.net/), [David F. Fouhey<sup>3</sup>](https://cs.nyu.edu/~fouhey/), [Chen-Hsuan Lin<sup>1</sup>](https://chenhsuanlin.bitbucket.io/) \
|
16 |
<sup>1</sup>NVIDIA <sup>2</sup>University of Michigan <sup>3</sup>New York University
|
17 |
|
|
|
|
|
|
|
|
|
18 |
[](https://research.nvidia.com/labs/dir/dynpose-100k) [](https://arxiv.org/abs/2504.17788)
|
19 |
|
20 |

|
@@ -122,11 +126,60 @@ dynpose_100k
|
|
122 |
| βββ ...
|
123 |
```
|
124 |
|
125 |
-
##
|
|
|
|
|
|
|
|
|
|
|
|
|
126 |
|
127 |
-
|
|
|
|
|
|
|
|
|
|
|
128 |
|
129 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
130 |
|
131 |
## FAQ
|
132 |
|
|
|
15 |
[Ming-Yu Liu<sup>1</sup>](https://mingyuliu.net/), [David F. Fouhey<sup>3</sup>](https://cs.nyu.edu/~fouhey/), [Chen-Hsuan Lin<sup>1</sup>](https://chenhsuanlin.bitbucket.io/) \
|
16 |
<sup>1</sup>NVIDIA <sup>2</sup>University of Michigan <sup>3</sup>New York University
|
17 |
|
18 |
+
## π Updates
|
19 |
+
- **[2025.05]** We have released the Lightspeed benchmark, a new dataset with ground-truth camera poses for validating DynPose-100K's pose annotation method. See [download instructions](#lightspeed-benchmark-download) below.
|
20 |
+
- **[2025.04]** We have made the initial release of DynPose-100K, a large-scale dataset of diverse, dynamic videos with camera annotations. See [download instructions](#dynpose-100k-download) below.
|
21 |
+
|
22 |
[](https://research.nvidia.com/labs/dir/dynpose-100k) [](https://arxiv.org/abs/2504.17788)
|
23 |
|
24 |

|
|
|
126 |
| βββ ...
|
127 |
```
|
128 |
|
129 |
+
## Lightspeed Benchmark Download
|
130 |
+
|
131 |
+
Lightspeed is a challenging, photorealistic benchmark for dynamic pose estimation with **ground-truth** camera poses. It is used to validate DynPose-100K's pose annotation method.
|
132 |
+
|
133 |
+
Original video clips can be found here: https://www.youtube.com/watch?v=AsykNkUMoNU&t=1s
|
134 |
+
|
135 |
+
### 1. Downloading cameras, videos and frames (8.1 GB)
|
136 |
|
137 |
+
```bash
|
138 |
+
git clone https://huggingface.co/datasets/nvidia/dynpose-100k
|
139 |
+
cd dynpose-100k
|
140 |
+
unzip lightspeed.zip
|
141 |
+
export LIGHTSPEED_PARENT=$(pwd)/lightspeed
|
142 |
+
```
|
143 |
|
144 |
+
### 2. Dataset structure
|
145 |
+
```
|
146 |
+
lightspeed
|
147 |
+
βββ poses.pkl
|
148 |
+
| βββ 0120_LOFT {id_setting}
|
149 |
+
| βββ poses {camera poses (all frames) ([N,3,4])}
|
150 |
+
| # where N is number of frames
|
151 |
+
| βββ 0180_DUST
|
152 |
+
| βββ poses {camera poses (all frames) ([N,3,4])}
|
153 |
+
| βββ ...
|
154 |
+
βββ video
|
155 |
+
| βββ 0120_LOFT.mp4 {id_setting}
|
156 |
+
| βββ 0180_DUST.mp4
|
157 |
+
| βββ ...
|
158 |
+
βββ frames-24fps
|
159 |
+
| βββ 0120_LOFT/images {id_setting}
|
160 |
+
| βββ 00000.png {frame id}
|
161 |
+
| βββ 00001.png
|
162 |
+
| βββ ...
|
163 |
+
| βββ 0180_DUST/images
|
164 |
+
| βββ 00000.png
|
165 |
+
| βββ ...
|
166 |
+
| βββ ...
|
167 |
+
βββ viz_list.txt {used as index for pose visualization}
|
168 |
+
| βββ 0120_LOFT.mp4 {id_setting}
|
169 |
+
| βββ 0180_DUST.mp4
|
170 |
+
| βββ ...
|
171 |
+
```
|
172 |
+
|
173 |
+
### 3. Camera pose visualization
|
174 |
+
Create a conda environment if you haven't done so:
|
175 |
+
```bash
|
176 |
+
conda env create -f environment.yml
|
177 |
+
conda activate dynpose-100k
|
178 |
+
```
|
179 |
+
Run the below under the `dynpose-100k` environment:
|
180 |
+
```bash
|
181 |
+
python visualize_pose.py --dset lightspeed --dset_parent ${LIGHTSPEED_PARENT}
|
182 |
+
```
|
183 |
|
184 |
## FAQ
|
185 |
|
lightspeed.zip
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:4e168f49f2a8facc8ed014868d0a7de408e6b6bedcc88067b821a88a182fb90c
|
3 |
+
size 8732366493
|