How neatherblok/snowdetection works
This research prototype prepares sidewalk-image datasets, trains snow/clear classifiers, evaluates individual and ensemble models, and targets mobile inference. The sampled T2T-ViT script also implements model layers, staged training, validation, and testing; its preparation and quantization links are visible in source. README documents synthetic augmentation, spatial attention, VGG-19/ResNet-50 ensembling, evaluation, and mobile use, but several corresponding notebooks were not sampled, so their internal wiring is not inferred.
Main entry points: TrainingT2T_Vit.py T2T-ViT classifier · TrainingVGG_RESNET_(ensemble).ipynb CNN ensemble · t2t-vit_quantization.py Mobile optimization · Preparation.py Image preparation
Components 17
Data preparation
- Image preparation
IPYNB/DataPreparation/Preparation.py - Synthetic augmentation
IPYNB - Prepared datasets
Model development
- T2T-ViT classifier
IPYNB/ModelCreation/TrainingT2T_Vit.py - Spatial attention CNN
IPYNB/ModelCreation/TrainingCNN_(using_spatial_attention).ipynb - VGG-19 model
IPYNB/ModelCreation/TrainingVGG_RESNET_(ensemble).ipynb - ResNet-50 model
IPYNB/ModelCreation/TrainingVGG_RESNET_(ensemble).ipynb - CNN ensemble
IPYNB/ModelCreation/TrainingVGG_RESNET_(ensemble).ipynb - Staged training
IPYNB/ModelCreation/TrainingT2T_Vit.py - Model checkpoints
models
Evaluation
- CNN evaluation
IPYNB/ModelEvaluation/CNN_Evaluation.ipynb - Ensemble evaluation
IPYNB/ModelEvaluation/Ensemble_Evaluation.ipynb - Classification metrics
IPYNB/ModelEvaluation
Mobile deployment
- Mobile optimization
IPYNB/ModelQuantization/t2t-vit_quantization.py - Mobile model artifact
IPYNB/ModelQuantization/t2t-vit_quantization.py
Other components
- Sidewalk images
- Pedestrian user
Connections 19
- Sidewalk images → Image preparation: provides images
IPYNB/DataPreparation/Preparation.py - Image preparation → Prepared datasets: loads and transforms
IPYNB/DataPreparation/Preparation.py - Synthetic augmentation → Prepared datasets: augments data
README.md - Prepared datasets → Staged training: trains with
IPYNB/ModelCreation/TrainingT2T_Vit.py - Staged training → T2T-ViT classifier: optimizes
IPYNB/ModelCreation/TrainingT2T_Vit.py - Staged training → Model checkpoints: saves weights
IPYNB/ModelCreation/TrainingT2T_Vit.py - Prepared datasets → Spatial attention CNN: trains with
README.md - Prepared datasets → VGG-19 model: trains with
README.md - Prepared datasets → ResNet-50 model: trains with
README.md - VGG-19 model → CNN ensemble: contributes predictions
README.md - ResNet-50 model → CNN ensemble: contributes predictions
README.md - T2T-ViT classifier → CNN evaluation: evaluated by
README.md - CNN ensemble → Ensemble evaluation: evaluated by
README.md - CNN evaluation → Classification metrics: reports metrics
README.md - Ensemble evaluation → Classification metrics: reports metrics
README.md - T2T-ViT classifier → Mobile optimization: imports model
IPYNB/ModelQuantization/t2t-vit_quantization.py - Model checkpoints → Mobile optimization: loads weights
IPYNB/ModelQuantization/t2t-vit_quantization.py - Mobile optimization → Mobile model artifact: exports optimized model
IPYNB/ModelQuantization/t2t-vit_quantization.py - Mobile model artifact → Pedestrian user: supports alerts
README.md