Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,38 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: mit
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
datasets:
|
| 4 |
+
- allenai/objaverse
|
| 5 |
+
- allenai/objaverse-xl
|
| 6 |
+
- qizekun/OrienText300K
|
| 7 |
+
- tiange/Cap3D
|
| 8 |
+
language:
|
| 9 |
+
- en
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
|
| 13 |
+
# PointSO
|
| 14 |
+
|
| 15 |
+
We introduce the concept of semantic orientation, representing the object orientation condition on open vocabulary language.
|
| 16 |
+
The model is part of [SoFar](https://github.com/qizekun/SoFar), a 6-DoF system for spatial reasoning and robotic manipulation.
|
| 17 |
+
The released weights is on [Huggingface PointSO](https://huggingface.co/qizekun/PointSO), and the code is in the [orientation](./orientation) folder.
|
| 18 |
+
|
| 19 |
+
## Pretrain
|
| 20 |
+
Download the PointMAE as initialization.
|
| 21 |
+
```bash
|
| 22 |
+
wget https://github.com/Pang-Yatian/Point-MAE/releases/download/main/pretrain.pth -P orientation/
|
| 23 |
+
```
|
| 24 |
+
Perpare the OrienText300K dataset following [DATASET.md](./datasets/DATASET.md).
|
| 25 |
+
```bash
|
| 26 |
+
cd orientation
|
| 27 |
+
sh train_ddp.sh
|
| 28 |
+
```
|
| 29 |
+
|
| 30 |
+
## Finetune
|
| 31 |
+
Perpare the Open6DOR finetuning dataset following [DATASET.md](./datasets/DATASET.md).
|
| 32 |
+
The dataset is generated from isaac sim with different assets from Open6DOR.
|
| 33 |
+
Finetune PointSO will significantly improve the performance on Open6DOR rotation track & 6-DoF track.
|
| 34 |
+
We recommend using this version of the model for the Open6DOR V2 evaluation.
|
| 35 |
+
```bash
|
| 36 |
+
cd orientation
|
| 37 |
+
sh train_ddp_ft.sh
|
| 38 |
+
```
|