Update README.md
Browse files
README.md
CHANGED
|
@@ -262,6 +262,89 @@ huggingface-cli download lightx2v/Autoencoders \
|
|
| 262 |
--local-dir ./models/vae/
|
| 263 |
```
|
| 264 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 265 |
### Use in LightX2V
|
| 266 |
|
| 267 |
Specify the VAE path in the configuration file:
|
|
|
|
| 262 |
--local-dir ./models/vae/
|
| 263 |
```
|
| 264 |
|
| 265 |
+
### 🧪 Video Reconstruction Test
|
| 266 |
+
|
| 267 |
+
We provide a standalone script `vid_recon.py` to test VAE models independently. This script reads a video, encodes it through VAE, then decodes it back to verify the reconstruction quality.
|
| 268 |
+
|
| 269 |
+
**Script Location**: `LightX2V/lightx2v/models/video_encoders/hf/vid_recon.py`
|
| 270 |
+
|
| 271 |
+
```bash
|
| 272 |
+
git clone https://github.com/ModelTC/LightX2V.git
|
| 273 |
+
cd LightX2V
|
| 274 |
+
```
|
| 275 |
+
|
| 276 |
+
**1. Test Official VAE (Wan2.1)**
|
| 277 |
+
```bash
|
| 278 |
+
python -m lightx2v.models.video_encoders.hf.vid_recon \
|
| 279 |
+
input_video.mp4 \
|
| 280 |
+
--checkpoint ./models/vae/Wan2.1_VAE.pth \
|
| 281 |
+
--model_type vaew2_1 \
|
| 282 |
+
--device cuda \
|
| 283 |
+
--dtype bfloat16
|
| 284 |
+
```
|
| 285 |
+
|
| 286 |
+
**2. Test Official VAE (Wan2.2)**
|
| 287 |
+
```bash
|
| 288 |
+
python -m lightx2v.models.video_encoders.hf.vid_recon \
|
| 289 |
+
input_video.mp4 \
|
| 290 |
+
--checkpoint ./models/vae/Wan2.2_VAE.pth \
|
| 291 |
+
--model_type vaew2_2 \
|
| 292 |
+
--device cuda \
|
| 293 |
+
--dtype bfloat16
|
| 294 |
+
```
|
| 295 |
+
|
| 296 |
+
**3. Test LightTAE (Wan2.1)**
|
| 297 |
+
```bash
|
| 298 |
+
python -m lightx2v.models.video_encoders.hf.vid_recon \
|
| 299 |
+
input_video.mp4 \
|
| 300 |
+
--checkpoint ./models/vae/lighttaew2_1.pth \
|
| 301 |
+
--model_type taew2_1 \
|
| 302 |
+
--device cuda \
|
| 303 |
+
--dtype bfloat16
|
| 304 |
+
```
|
| 305 |
+
|
| 306 |
+
**4. Test LightTAE (Wan2.2)**
|
| 307 |
+
```bash
|
| 308 |
+
python -m lightx2v.models.video_encoders.hf.vid_recon \
|
| 309 |
+
input_video.mp4 \
|
| 310 |
+
--checkpoint ./models/vae/lighttaew2_2.pth \
|
| 311 |
+
--model_type taew2_2 \
|
| 312 |
+
--device cuda \
|
| 313 |
+
--dtype bfloat16
|
| 314 |
+
```
|
| 315 |
+
|
| 316 |
+
**5. Test LightVAE (Wan2.1)**
|
| 317 |
+
```bash
|
| 318 |
+
python -m lightx2v.models.video_encoders.hf.vid_recon \
|
| 319 |
+
input_video.mp4 \
|
| 320 |
+
--checkpoint ./models/vae/lightvaew2_1.pth \
|
| 321 |
+
--model_type vaew2_1 \
|
| 322 |
+
--device cuda \
|
| 323 |
+
--dtype bfloat16 \
|
| 324 |
+
--use_lightvae
|
| 325 |
+
```
|
| 326 |
+
|
| 327 |
+
|
| 328 |
+
**6. Test TAE (Wan2.1)**
|
| 329 |
+
```bash
|
| 330 |
+
python -m lightx2v.models.video_encoders.hf.vid_recon \
|
| 331 |
+
input_video.mp4 \
|
| 332 |
+
--checkpoint ./models/vae/taew2_1.pth \
|
| 333 |
+
--model_type taew2_1 \
|
| 334 |
+
--device cuda \
|
| 335 |
+
--dtype bfloat16
|
| 336 |
+
```
|
| 337 |
+
|
| 338 |
+
**7. Test TAE (Wan2.2)**
|
| 339 |
+
```bash
|
| 340 |
+
python -m lightx2v.models.video_encoders.hf.vid_recon \
|
| 341 |
+
input_video.mp4 \
|
| 342 |
+
--checkpoint ./models/vae/taew2_2.pth \
|
| 343 |
+
--model_type taew2_1 \
|
| 344 |
+
--device cuda \
|
| 345 |
+
--dtype bfloat16
|
| 346 |
+
```
|
| 347 |
+
|
| 348 |
### Use in LightX2V
|
| 349 |
|
| 350 |
Specify the VAE path in the configuration file:
|