Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language: en
|
| 3 |
+
tags:
|
| 4 |
+
- point-cloud
|
| 5 |
+
- 3d-classification
|
| 6 |
+
- pytorch
|
| 7 |
+
- pointnet
|
| 8 |
+
- modelnet10
|
| 9 |
+
license: mit
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
# Vanilla PointNet — ModelNet10 Classifier
|
| 13 |
+
|
| 14 |
+
Simplified PointNet (no T-Nets) for 3D object classification on ModelNet10.
|
| 15 |
+
|
| 16 |
+
## Architecture
|
| 17 |
+
| Stage | Layers |
|
| 18 |
+
|-------|--------|
|
| 19 |
+
| Feature Extraction | Conv1d(3→64→128→1024) + BN + ReLU |
|
| 20 |
+
| Global Aggregation | Global max-pool → 1024-d vector |
|
| 21 |
+
| Classification | FC(1024→512→256→10) + BN + Dropout |
|
| 22 |
+
|
| 23 |
+
## Classes
|
| 24 |
+
`bathtub · bed · chair · desk · dresser · monitor · night_stand · sofa · table · toilet`
|
| 25 |
+
|
| 26 |
+
## Usage
|
| 27 |
+
```python
|
| 28 |
+
import torch
|
| 29 |
+
from model import VanillaPointNet
|
| 30 |
+
|
| 31 |
+
model = VanillaPointNet(num_classes=10)
|
| 32 |
+
model.load_state_dict(torch.load("pytorch_model.bin", map_location="cpu"))
|
| 33 |
+
model.eval()
|
| 34 |
+
|
| 35 |
+
pts = torch.randn(1, 3, 1024) # (B, 3, N)
|
| 36 |
+
print(model(pts).argmax(1))
|
| 37 |
+
```
|