How maaxxe/train_kaggle_plank_detector works
This repository is a collection of computer-vision training projects rather than a single deployed application. Its main workflow is annotated plank images → dataset preparation and augmentation → quadrilateral detector training → model checkpoints; a separate GAN project trains synthetic-image generation, while GeoNet targets segmentation and instance extraction. The tree also contains a PaddlePaddle GeoNet port and checkpoint-conversion path. README contents were not provided, and excerpts are partial, so unsampled subsystem wiring is shown only where the tree or sampled code supports it.
Main entry points: train_v7.py Detector training · model_v7_1.py Quadrilateral detector · models.py Generator and discriminator · train.py GAN training
Components 19
PlankEye detection
- Dataset and splits
Plankeye/train/train_v7.py - Image augmentation
Plankeye/train/train_v7.py - Quadrilateral detector
Plankeye/model/model_v7_1.py - Detector training
Plankeye/train/train_v7.py - Detector checkpoints
Plankeye/model_poids - Corner predictions
Plankeye/model/model_v7_1.py
Synthetic data GAN
- Mask image pairs
GAN_PlankEYE/dataset.py - Generator and discriminator
GAN_PlankEYE/models.py - GAN training
GAN_PlankEYE/train.py - Synthetic images
GAN_PlankEYE/generate.py
GeoNet segmentation
- Images and labels
GeoNet_paddle/train/train_geonet_baidu.py - MultiFormeNet
GeoNet/model/multiforme_model.py - GeoNet training
GeoNet_paddle/train/train_geonet_baidu.py - Instance extraction
GeoNet/model/multiforme_model.py
Paddle port
- PyTorch checkpoint
GeoNet_paddle/tools/export_torch_checkpoint_npz.py - NPZ weights
GeoNet_paddle/model_poids - Paddle MultiFormeNet
GeoNet_paddle/model/multiforme_model.py - Paddle training
GeoNet_paddle/train/train_geonet_baidu.py
Other components
- Annotated images
Connections 15
- Annotated images → Dataset and splits: provides labels
- Dataset and splits → Image augmentation: loads samples
- Detector training → Dataset and splits: builds loaders
- Detector training → Quadrilateral detector: trains
- Detector training → Detector checkpoints: saves weights
- Quadrilateral detector → Corner predictions: predicts corners
- Mask image pairs → GAN training: provides pairs
- GAN training → Generator and discriminator: trains
- Generator and discriminator → Synthetic images: generates
- Images and labels → GeoNet training: provides samples
- GeoNet training → MultiFormeNet: trains
- MultiFormeNet → Instance extraction: extracts instances
- PyTorch checkpoint → NPZ weights: converts to
- NPZ weights → Paddle training: initializes from
- Paddle training → Paddle MultiFormeNet: trains