Add dataset card

README.md

@@ -21,7 +21,7 @@ size_categories:
 # Dataset Overview
 
 ## Dataset Description:
-We are releasing 50 sample images of components on NV boards in a FoxConn factory. These images are used to finetune the Cosmos Transfer 2.5 models and Qwen-Image-Edit.
+This dataset contains printed-circuit-board (PCB) component crops captured under raked solder-light illumination, organized for PCB anomaly generation and inspection. Images are grouped by board region — `IC` and `passive_component` — and for each region include anomaly images (defect types `bridge`, `excess_solder`, `missing`), their paired binary defect masks, clean reference images, and CAD-derived masks, along with a `defect_spec.jsonl` specification and `semantic_segmentation_labels.json`. The set is used to fine-tune the Cosmos AnomalyGen PCB model for synthetic PCB defect-data generation.
 
 This dataset is for demonstration purposes and not for production usage.<br>
 
@@ -29,7 +29,7 @@ This dataset is for demonstration purposes and not for production usage.<br>
 NVIDIA Corporation<br>
 
 ## Dataset Creation Date:
-05/01/2025<br>
+05/29/2026<br>
 
 ## Version:
 1.0<br>
@@ -42,16 +42,21 @@ Demonstration of software to generate synthetic data of structural defects in a
 
 ## Dataset Characterization
 **Data Collection Method by Dataset:**
-- Human
+- Hybrid — Human, Automatic/Sensors (photographic capture at PCB inspection stations, plus synthetic image augmentations such as flips, rotations, and brightness adjustments)
 
 **Labeling Method by Dataset:**
-- Not Applicable<br>
+- Human (per-defect binary masks)<br>
 
 ## Dataset Format
-PNG / JPG files.<br>
+PNG / JPG image files (with accompanying JSON / JSONL metadata).<br>
 
 ## Dataset Quantification
-50 sample images, 6.62 GB.<br>
+182 image files (~2 MB total), organized under `IC/` and `passive_component/`:
+- 86 anomaly images — `bridge` (8), `excess_solder` (16), `missing` (62)
+- 86 paired binary defect masks
+- 5 clean reference images and 5 CAD-derived masks
+
+Approximately 108 of the 182 images are synthetic augmentations (horizontal/vertical flips, rotations, brightness adjustments) of the base captures. Accompanying metadata: `defect_spec.jsonl` and `semantic_segmentation_labels.json`.<br>
 
 ## Ethical Considerations:
 NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. Developers should work with their internal developer teams to ensure this dataset meets requirements for the relevant industry and use case and addresses unforeseen product misuse.<br>