AutoLens EfficientNet-B2 Candidate
Current strongest CNN candidate for the AutoLens AI demo artifact lane.
Model
- Architecture: EfficientNet-B2
- Classes: 8 vehicle body types
- Source run:
baseline_0_efficientnet_b2_20260509_135313
- Source checkpoint:
checkpoints/baseline_0_efficientnet_b2_20260509_135313/best-04-0.8994.ckpt
- Export format:
model.safetensors plus metadata.json, with ONNX for deployment
Classes
- SUV
- VAN
- STATION WAGON
- MICRO
- OPEN WHEEL / F1
- SEDAN
- HATCHBACK
- PICK UP
Preprocessing
- RGB input
- Resize: 256
- Center crop / model input: 224 x 224
- Mean:
[0.4429, 0.4354, 0.437]
- Std:
[0.2456, 0.2421, 0.2449]
Internal Test Results
Evaluated on held-out internal test set (3170 samples, no data leakage).
- Accuracy: 0.9202
- F1-macro: 0.8982
- F1-weighted: 0.9201
- Precision-macro: 0.9056
- Recall-macro: 0.8923
Per-class Performance
| Class | Precision | Recall | F1-score | Support |
|---|
| SUV | 0.9213 | 0.9090 | 0.9151 | 670 |
| VAN | 0.9450 | 0.9824 | 0.9634 | 455 |
| STATION WAGON | 0.8033 | 0.7424 | 0.7717 | 66 |
| MICRO | 0.9545 | 0.9545 | 0.9545 | 22 |
| OPEN WHEEL / F1 | 0.9948 | 0.9846 | 0.9897 | 586 |
| SEDAN | 0.9023 | 0.9282 | 0.9151 | 836 |
| HATCHBACK | 0.7662 | 0.8008 | 0.7831 | 266 |
| PICK UP | 0.9574 | 0.8364 | 0.8929 | 269 |
Temperature Scaling (Calibration)
Post-training calibration uses temperature scaling. Temperature was fit on the validation set only; the before/after values below are measured on the held-out internal test set.
- Temperature: 1.7799
- Internal test ECE before: 0.0393 → after: 0.0107
- Internal test NLL before: 0.2745 → after: 0.2301
- Internal test mean confidence before: 0.9595 → after: 0.9168
- Internal test accuracy: 0.9202 (3170 samples)
Calibration parameters are stored in calibration.json. The ONNX model outputs raw logits; apply temperature scaling at inference time.
Artifact sizes
model.safetensors: 29.722 MBmodel.onnx: 29.371 MB
Files
model.safetensors — weights-only model artifactmetadata.json — architecture, classes, preprocessing, source run, and artifact metadatamodel.onnx — ONNX Runtime inference artifactmodel.simplified.onnx — optional simplified ONNX graph when availablesize_report.json / size_check.json — artifact size and ONNX smoke evidence
Intended use
Educational demo and report evidence for classifying uploaded vehicle images into the 8 AutoLens body-type classes.
Limitations
The dataset is assembled from public/open sources and may contain domain bias. Similar body styles such as hatchback, station wagon, and sedan can be ambiguous. The merged raw dataset is not redistributed in this model repository.
