|
Download EXPLAINABILITY.md from nvidia/Cosmos-AnomalyGen-PCB-Nano: direct link, hf CLI and curl.
- Browser
- Download file 5.53 kB
-
https://huggingface.co/nvidia/Cosmos-AnomalyGen-PCB-Nano/resolve/main/EXPLAINABILITY.md
- Command line
-
hf download hf://nvidia/Cosmos-AnomalyGen-PCB-Nano/EXPLAINABILITY.md
-
curl -L -o EXPLAINABILITY.md https://huggingface.co/nvidia/Cosmos-AnomalyGen-PCB-Nano/resolve/main/EXPLAINABILITY.md
5.53 kB
Explainability
| Field | Response |
|---|---|
| Intended Task/Domain: | Industrial visual inspection — synthetic anomaly image generation (image-to-image inpainting) for printed-circuit-board defect detection. |
| Model Type: | Diffusion model — Transformer-based MoT denoiser (Cosmos3-Nano, frozen) conditioned by trained per-defect-type LoRA adapters on its generation-expert attention projections, with a frozen Wan2.2 VAE tokenizer and a frozen Qwen3-VL text encoder. |
| Intended Users: | ML engineers and computer-vision practitioners building defect-detection or segmentation models for printed-circuit-board QA. |
| Output: | Image (single RGB anomaly image at the clean image's native resolution per inference call; multiple per testcase line when num_generated_images > 1). |
| Describe how the model works: | A defect-centred crop of the clean reference image, with the user-supplied binary mask region replaced by noise, is encoded by the frozen Wan2.2 VAE and fed to the frozen Cosmos3-Nano MoT denoiser together with an editing caption built from the selected <texture>+<anomaly_type> (IC+bridge, passive_component+excess_solder, or passive_component+missing), whose generation tokens are routed to the trained LoRA adapter of that defect type. The denoiser then produces an inpainting-style sample where pixels inside the mask are generated as a defect that matches the trained type, while pixels outside the mask are preserved from the clean reference image (with crop-and-paste / optional Poisson blending). The two-texture design (IC vs. passive_component) lets a single checkpoint cover defects that appear differently depending on board region. |
| List the groups (or group characteristics) for which this was tested to produce comparable outcomes: | Not Applicable |
| Technical Limitations & Mitigation: | Trained on a small number of examples per defect pair (8, 16, and 62 anomaly images for IC+bridge, passive_component+excess_solder, and passive_component+missing respectively). Out-of-distribution board layouts, mask placements that deviate substantially from the trained distribution, mismatched texture choice (e.g., using IC+bridge on a passive-component region of the board), or any defect not in the three trained pairs may yield low-fidelity outputs. The IC+bridge pair has the fewest training examples and is the most sensitive to mask-placement quality. Mitigation: (a) use the Automatic Mask Placement (AMP) tool or the Agentic Mask Generation (AMG) stage, which automatically restricts the user-supplied mask to legal regions of interest for each <texture>+<anomaly_type> pair (e.g., only on IC pads for IC+bridge, only on passive-component pads for passive_component+*), preventing the model from generating defects in implausible board locations; (b) run anomalygen/scripts/texture/filter.py on the generated dataset — the filter scores each output by nearest-neighbour correspondence to the real defect examples (optionally combined with the anomaly-quality metrics) and discards the lowest-scoring fraction per defect type, so unrealistic synthetic images do not pollute the downstream training set; (c) validate any downstream detector trained on synthetic samples against real defect imagery before deployment. |
| Verified to have met prescribed NVIDIA quality standards: | Yes. |
| Performance Metrics: | FID (logged during training validation), nearest-neighbor metrics (nn_score, mnn_score), anomaly-quality metrics (completeness, precision, boundary_iou), and visual inspection of ValidationKPI callback outputs. |
| Potential Known Risks: | Generated synthetic defects may not perfectly cover all real-world defect variations. Downstream detection models trained primarily on these synthetic samples should be validated against real defect data before deployment in production QA lines. This model can generate synthetic images and may produce content that is offensive, unsafe, misleading, indecent, or unsuitable for a target deployment. Users should implement robust safety guardrails — including content filtering, abuse monitoring, and access controls — to reduce the risk of harmful outputs. Users are responsible for ensuring that their use of the model complies with all applicable laws and regulations, and for regularly reviewing and updating their guardrails as risks evolve. |
| Licensing: | OpenMDW License Agreement, version 1.1 |