How wll199566/learning-to-generate-chairs-with-convolutional-neural-networks works
This repository implements a PyTorch chair-image generator based on the cited 2014 paper. Its main workflow loads chair renders, masks, and class labels; encodes class and view conditions; trains a network to produce an image and segmentation mask; and validates and saves model checkpoints. A separate evaluation script loads a checkpoint and generates a result for requested conditions. The training excerpt is partial, so links for setup details not shown there are omitted; the dataset and evaluation behavior are supported by sampled source.
Main entry points: train.py Train and validate · evaluate.py Generate prediction · model.py Conditional generator · Dataset.py Load and encode
Components 9
Chair data
- Load and encode
Dataset.py
Model training
- Training settings
config.yaml - Train and validate
train.py - Conditional generator
modules/model.py - Model checkpoint
checkpoints
Generation and evaluation
- Generate prediction
evaluate.py - Image and mask
Other components
- Researcher
- Chair files
Connections 12
- Researcher → Train and validate: starts training
- Chair files → Load and encode: loads renders and labels
- Load and encode → Train and validate: supplies batches
- Training settings → Train and validate: sets defaults
- Train and validate → Conditional generator: trains and validates
- Conditional generator → Train and validate: returns predictions
- Train and validate → Model checkpoint: saves weights
- Researcher → Generate prediction: requests conditions
- Model checkpoint → Generate prediction: loads weights
- Generate prediction → Conditional generator: runs inference
- Conditional generator → Generate prediction: returns predictions
- Generate prediction → Image and mask: produces image and mask