Spaces:
Sleeping
Sleeping
Commit History
-
artifacts: add EfficientNet and ResNet50 confusion matrices and training history plots
Priya Monisha91a21f7 -
results: Section 5 complete — all 4 CNNs trained, ResNet50 best at 65.5%
Priya MonishaSection 5 final results: - VGG16: 59.5% (architecture ceiling) - MobileNetV2: 56.7% (backbone lr=1e-5 too conservative) - EfficientNetB0: 58.9% (backbone lr=1e-5 too conservative) - ResNet50: 65.5% test / 63.6% val (data volume bottleneck, 1464 params/img) Pipeline improvements committed alongside results: - dataset.py: create_stratified_split() — random stratified resplit, no sequential bias - loader.py: quality gates 48px→80px, 2% area; compute_overlap_ratio; check_person_overlap - model_factory.py: unfreeze_efficientnet_phase2() added - 01_data_acquisition.py: chair person-overlap filter, MAX_ITER 80k→150k Next: Section 6 YOLOv8 detection training1a28594
-
section-4: preprocessing + augmentation — verified end-to-end
Priya Monishasrc/data/preprocessor.py: - validate_yolo_annotations(): 1550 train + 375 val files, 13758 annotations all valid Fixed: assert <= 1.0 not < 1.0 (bbox_to_yolo clamps to min(1.0,...) so boundary values valid) - create_yolo_data_yaml(): absolute paths, works in terminal + Colab - verify_dataset_structure(): counts per split, detects image/label mismatches src/data/augmentor.py: - get_train_transforms(): flip, rotate±15, colorjitter, zoomout, resize 224, ToImage, Normalize(ImageNet) - get_eval_transforms(): resize + normalize only - denormalize(): for visualization - torch wrapped in try/except (local fbgemm.dll issue) notebooks/03_preprocessing.py: - Dataset verified: 2500 classification, 1925 detection, labels match images - YOLO annotations validated: 13758 annotations, all in valid range - data.yaml regenerated with current absolute paths - Augmented sample images saved to docs/figures/ - Tensor shapes confirmed: (3, 224, 224) float32, range ~[-2.1, 2.6] after ImageNet norm2ea8a2d -
section-3: EDA complete and verified — all 7 figures correct
Priya MonishaFixes applied after chart review: - Chart 3 (Aspect Ratio): was scanning 224x224 crops (all ratio=1.0, meaningless) Fixed to scan full detection images → real distribution, mean=1.29, landscape-dominant - Chart 7 (Difficulty): was predicting 21/25 classes as hard (too coarse supercategory logic) Fixed to visually-similar-pair logic → 6 hard (car/bus/truck/cup/chair/couch), sensible spread - Chi-squared note: added FAST_MODE caveat (chi2=0 trivially true with 10/class) - All 7 figures confirmed correct in docs/figures/4973d81