Educational prototype | Binary image classification | Human-reviewed inspection only
Model overview
| Model | Task | Framework | Weights |
|---|---|---|---|
| FactoryVision AI Custom CNN | Casting defect classification | PyTorch | model.pth |
Held-out test results
| Metric | Reported result |
|---|---|
| Accuracy | 0.9860 |
| Macro precision | 0.9849 |
| Macro recall | 0.9849 |
| Macro F1 | 0.9849 |
| Mean inference latency | 0.608 ms |
Confusion matrix — rows are true labels, columns are predictions; class order: def_front, ok_front.
[[448, 5],
[ 5, 257]]Metrics and latency are reported from the supplied Kaggle output, not independently reproduced. Latency is environment/measurement specific and is not a target CPU benchmark. The model is not validated on new factory deployments.
Purpose and limitations
Educational two-class classification of casting images (def_front = defective; ok_front = acceptable). For human-reviewed prototype inspection only; not an industrial release decision or certified quality system.
Model details
- Custom PyTorch CNN trained from scratch; see
src/quality.pyin the source repository for architecture and preprocessing - Packaged weights:
model.pth(archive pathcheckpoints/casting_cnn.pth) - Exact ordered labels:
labels.json - Use the project's original loader and preprocessing rather than assuming a generic Hugging Face Transformer loader can use the
.pthfile.
Training data
Real-Life Industrial Dataset of Casting Product by Ravirajsinh Dabhi (Kaggle dataset). The source README identifies its Kaggle licence as CC BY-NC-ND 4.0, including non-commercial and no-derivatives restrictions. Rights to publish dataset-trained checkpoint weights require separate verification; model-developer consent does not settle third-party licensing. No input images are redistributed here.
Inference and dependencies
The checkpoint is intended to be loaded using the matching architecture from src/quality.py in the source repository. Do not call arbitrary torch.load on untrusted checkpoints. The project has known image-size, normalisation, and class-order requirements; check the current source for the exact implementation before inference.
Licensing and distribution
The source project has an MIT licence for its code, not necessarily the checkpoint. Keep this repository private until the casting dataset's non-commercial/no-derivatives conditions and trained-weight publication rights have been resolved. No model-weight licence is asserted in this card.
Credits and attribution
| Role | Contributor |
|---|---|
| Original Developer & Model Author | Theivanth Rajamohan |
| Technical Reviewer, Code Contributor & Publication Support | Swapin Vidya |
| Project Contributor | Rajah Shanmugasundaram |
| Publishing Organisation | TechSphere Consultancy Services Private Limited |
Project provenance: FactoryVision AI was originally developed by Theivanth Rajamohan. Swapin Vidya contributed technical review, detection-metrics improvements, testing, model publication preparation and documentation. The models are being prepared for publication under TechSphere Consultancy Services Private Limited's Hugging Face namespace, with the original developer's permission.