Contamination Map Methodology

Vision / Model

Segment Anything Model (SAM)

Organisation
Meta AI
Family
Segment Anything · version 1
Released
2023-04-05 (arXiv v1 submission date (https://arxiv.org/abs/2304.02643v1))
Checkpoints
sam_vit_h_4b8939.pth (vit_h), sam_vit_l_0b3195.pth (vit_l), vit_b
Weights
Open · Apache-2.0
Links
PaperCode

Training data disclosure: Partial list

Level determined for 25% of benchmark cells.

“SAM was trained on SA-1B”
arxiv.org · paper · Section 7 (Implementation) · retrieved 2026-10-15
“We initialize SAM from an MAE [47] pre-trained ViT-H.”
arxiv.org · paper · Appendix A (Training algorithm) · retrieved 2026-10-15

Training corpora

Benchmark cells

Industrial anomaly detection

Unknown MVTec AD · training data not disclosed in enough detail

Not seen MVTec AD 2 · its test items were published from 2025-03-27, after the model's stated training-data cutoff (2022-12-31)

Evidence (3)
“The licensed photos vary in their date taken over a wide range of years up to 2022.”
arxiv.org · paper · Datasheet (Collection Process, Q5) · retrieved 2026-10-15
“arXiv:2503.21622v1 [cs.CV] 27 Mar 2025”
arxiv.org · paper · header · retrieved 2026-10-15
“All image data is available at https://www.mvtec.com/company/research/datasets/mvtec-ad-2.”
arxiv.org · paper · Abstract · retrieved 2026-10-15

Unknown MVTec LOCO AD · training data not disclosed in enough detail

Unknown VisA · training data not disclosed in enough detail

Not seen Real-IAD · its test items were published from 2024-03-19, after the model's stated training-data cutoff (2022-12-31)

Evidence (2)
“The licensed photos vary in their date taken over a wide range of years up to 2022.”
arxiv.org · paper · Datasheet (Collection Process, Q5) · retrieved 2026-10-15
“arXiv:2403.12580v1 [cs.CV] 19 Mar 2024”
arxiv.org · paper · header · retrieved 2026-10-15

Unknown MPDD · training data not disclosed in enough detail

Unknown BTAD · training data not disclosed in enough detail

Unknown KolektorSDD2 · training data not disclosed in enough detail

Notes

The image encoder is initialised from an MAE pre-trained ViT-H. The SAM paper does not state the MAE pre-training data. The MAE paper (https://arxiv.org/pdf/2111.06377v3, Section 4) says 'We do self-supervised pre-training on the ImageNet-1K (IN1K) training set', but SAM does not say which MAE checkpoint it used. disclosure is partial_list for that reason. In the data engine's first stage, SAM was first trained on common public segmentation datasets, then 'retrained using only newly annotated masks' (Section 4). The released model is described as trained on SA-1B.

Sources