ISL Unconditional Diffusion
An unconditional DDPM trained from scratch to generate 128×128 RGB images of Indian Sign Language hand gestures.
The model learns the overall distribution of ISL gesture images without receiving a class label during training or inference.
The complete implementation and experiments are available in the GitHub repository.
Model
- DDPM with UNet2D architecture
- 128×128 resolution
- RGB images
- Unconditional generation
- EMA weights
Training
- Dataset: 42,000 images (1,200 images per class × 35 classes)
- Noise schedule: cosine
- Batch size: 64
- Learning rate: 1e-4
- Mixed precision: fp16
- Training steps: 65,000
- EMA decay: 0.9999
- Data augmentation: enabled
Sampling
- Default sampler: DDIM
- Training diffusion timesteps: 1,000
- Routine inference steps: 100
- Evaluation inference steps: 50
- Random seed: 42
Results
The unconditional model achieves an FID of 86.06 under the reported evaluation setting.