DetectiveSAMv2

DetectiveSAMv2 is an inference-only image forgery localization bundle built around SAM2. This repo is v2-only: the code path, config sidecar, tests, and demo commands all target the detective_sam_v2 checkpoint.

The full runnable bundle with model weights is hosted on Hugging Face:

• https://huggingface.co/Gertlek/DetectiveSAMv2

What is bundled

• Inference checkpoint configs under checkpoints/
• SAM2 config under sam2configs/
• Poster demo pairs under demo/cocoglide/, demo/flux_test/, and demo/qwen_test/
• A drop-in single-image slot at demo/user_image/demo_input.png

Built-in checkpoint alias:

• detective_sam_v2

The detective_sam_v2 alias expects checkpoints/detective_sam_v2.pth and the bundled YAML sidecar.

DetectiveSAMv2 Release

This release packages the v2 architecture and benchmarked weights without the old DetectiveSAMv1 compatibility branches.

Main changes:

• Evidence features built from unadapted target features, raw perturbation-stream features, and adapted target deltas.
• Spatial cross-attention feature adapters for flexible information sharing across target and perturbation streams.
• Transformer evidence mask adapter for direct prompt-mask prediction.
• V2-only adapter implementation in detectivesam_inference/models/adapters.py.
• Three perturbation streams by default: Gaussian blur, JPEG compression, and Gaussian noise.
• Training scaled beyond the original SIDA/MagicBrush mix with PicoBanana 10k clean, UltraEdit 10k, and AutoEditForge-generated QWEN/FLUX train data designed to promote generalization to new editors.
• Validation expanded across trained-on and held-out modern edit benchmarks, including QWEN-Bench, FLUX-Bench, CoCoGLIDE, AutoSplice, and NanoBanana.
• JSON and YAML checkpoint sidecar support, including training checkpoint containers with model or ema keys.

Benchmark Scores

Scores below are IoU / F1 in percent for the 10-epoch v2 checkpoint dsamv2_best6info_10ep_20260623_060047/best_model.pth. Paper baseline numbers are from the Detective SAM ICLR 2026 paper tables for the original Detective SAM model.

Dataset	Split	DetectiveSAMv2	Paper Detective SAM	Delta
MagicBrush	trained on	62.73 / 74.38	46.48 / 57.55	+16.25 / +16.83
SIDA	trained on	55.34 / 65.72	54.55 / 65.29	+0.79 / +0.43
UltraEdit 10k	trained on	67.03 / 76.71	27.74 / 35.54	+39.29 / +41.17
PicoBanana 10k clean	trained on	41.17 / 55.75	n/a	n/a
FLUX-Bench	trained on	46.10 / 58.02	18.70 / 21.28	+27.40 / +36.74
QWEN-Bench	trained on	46.32 / 57.94	20.41 / 22.29	+25.91 / +35.65
CoCoGLIDE	not trained on	47.17 / 58.83	44.74 / 51.50	+2.43 / +7.33
AutoSplice	not trained on	58.39 / 71.40	46.90 / 60.30	+11.49 / +11.10
NanoBanana	not trained on	33.81 / 46.79	19.34 / 20.77	+14.47 / +26.02

PicoBanana 10k clean is the curated training-set split evaluated with the inference package at threshold 0.5; it has no Detective SAM paper baseline. QWEN-Bench and FLUX-Bench are reported as trained-on because the training mix includes corresponding AutoEditForge-generated QWEN and FLUX datasets.

Grouped results using the same checkpoint:

Group	Weighted IoU / F1	Simple IoU / F1
Trained-on	54.33 / 66.01	53.12 / 64.75
Not-trained-on	56.54 / 69.39	46.46 / 59.01
Overall	54.61 / 66.45	50.90 / 62.84

The not-trained-on group follows the requested reporting split and includes NanoBanana, CoCoGLIDE, and AutoSplice.

Setup

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Then download the large weights from Hugging Face:

pip install -U huggingface_hub
hf download Gertlek/DetectiveSAMv2 \
  checkpoints/detective_sam_v2.pth \
  sam2configs/sam2.1_hiera_base_plus.pt \
  --local-dir .

The expected checkpoint paths are also documented in checkpoints/README.md and sam2configs/README.md.

Hugging Face Usage

For the simplest setup, clone the Hugging Face repo directly:

git lfs install
git clone https://huggingface.co/Gertlek/DetectiveSAMv2
cd DetectiveSAMv2
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python -m detectivesam_inference.predict \
  --output-dir outputs/poster_baseline

Poster Demo Flows

1. Live single-image demo

Place your image at demo/user_image/demo_input.png, then run:

python -m detectivesam_inference.predict \
  --output-dir outputs/poster_user_image

In this mode the CLI reuses the target image as its own source reference so the demo stays runnable with a single image.

2. Bundled CocoGLIDE example

If demo/user_image/demo_input.png is absent, the default predict command falls back to the bundled CocoGlide sample banana_28809.

python -m detectivesam_inference.predict \
  --output-dir outputs/poster_baseline

3. Bundled modern-edit examples

Flux example:

python -m detectivesam_inference.predict \
  --checkpoint detective_sam_v2 \
  --source demo/flux_test/source/548.png \
  --target demo/flux_test/target/548.png \
  --mask demo/flux_test/mask/548.png \
  --output-dir outputs/poster_flux

Qwen example:

python -m detectivesam_inference.predict \
  --checkpoint detective_sam_v2 \
  --source demo/qwen_test/source/166.png \
  --target demo/qwen_test/target/166.png \
  --mask demo/qwen_test/mask/166.png \
  --output-dir outputs/poster_qwen

4. Bundled CocoGlide subset sweep

Use this to evaluate the bundled banana and train CocoGlide demo pairs.

python -m detectivesam_inference.evaluate \
  --checkpoint detective_sam_v2 \
  --dataset-root demo/cocoglide \
  --output-dir outputs/poster_eval_cocoglide \
  --num-visualizations 2

Outputs

Each predict run writes a compact set of visual artifacts plus a JSON summary:

• <name>_comparison.png
• <name>_probability.png
• <name>_pred_mask.png
• <name>_pred_overlay.png
• <name>_summary.json

If a ground-truth mask is provided, the run also saves:

• <name>_gt_mask.png
• <name>_gt_overlay.png

The evaluate command writes summary.json plus a few visualization examples under visualizations/.

Notes

• The runtime selects cuda automatically when available and otherwise runs on CPU.
• Checkpoint settings come from YAML or JSON sidecars in checkpoints/; you only need the alias or checkpoint path.
• The Hugging Face repo bundles the v2 .pth checkpoint and SAM2 .pt weight file.
• This repo does not include training code or training-only dependencies.