Synth.Eye GAN — Industrial Inspection Dataset Training dataset for the Synth.Eye GAN project — synthetic YOLO format images of industrial parts with and without fingerprint residue defects, generated by three StyleGAN2 ADA models and a physically motivated compositing pipeline. The dataset drives two downstream YOLO models: one for part orientation classification and one for fingerprint defect detection. Both models are available at LukasMoravansky/Synth Eye GAN. Dataset Structure Classes ID Name Description 0 Cls Obj Front Side Front side of the industrial part (no defect) 1 Cls Obj Back Side Back side of the industrial part 2 Cls Defect Fingerprint Fingerprint residue composite on the front side Splits Split Content Size train Synthetic GAN images (80 %) ~4 800 images val Synthetic GAN images (10 %) ~600 images test Real industrial photos from INTEMAC Research Center varies Default generation counts: 3 000 back side + 1 500 front side + 1 500 front+fingerprint composites = 6 000 synthetic images . Format Images: 256 × 256 px PNG, composited onto a green camera background Labels: YOLO .txt (one box per line: class cx cy w h , normalized), mirroring the image directory tree Config:…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy