DAMP (Detection-Anchored Multi-Scale Pooling), a new simple pooling algorithm for one-stage object detectors such as YOLO
SwiGLU gated MLP for all feed-forward blocks (½ the params of ReLU-MLP at equal capacity)
Visual x Semantic cross-attention — visual tokens attend to GloVe prototype embeddings
Geometry RoPE — box-position encoded as a rotary frequency bias on the Q matrix
Prototype Momentum Buffer — per-class EMA prototype bank
P5 Scene Context — AIFI-enhanced P5 tokens provide global context via cross-attention

The models were trained with the SGG-Benchmark framework and described in the REACT++ paper (Neau et al., 2026).

Results — SGDet on PSG test split (ONNX, CUDA)

Metrics from end-to-end ONNX evaluation (tools/eval_onnx_psg.py). E2E Latency = image load + pre-process + ONNX forward.

Backbone	Params	R@20	R@50	R@100	mR@20	mR@50	mR@100	F1@20	F1@50	F1@100	E2E Lat. (ms)
yolo12l	~26.5M	30.99	35.3	36.68	23.2	25.49	26.45	26.53	29.6	30.74	19.6
yolo12m	~20.2M	32.69	37.2	38.58	22.74	25.21	26.08	26.82	30.05	31.12	15.7
yolo12s	~9.2M	29.28	33.48	34.74	21.12	23.21	23.77	24.54	27.41	28.23	12.2
yolo12n	~2.6M	26.88	30.61	31.8	16.88	18.65	19.5	20.74	23.17	24.17	11.4
yolov8m	~25.9M	30.69	35.68	37.43	22.75	25.46	26.4	26.13	29.72	30.96	15.3

Checkpoints

Variant	Sub-folder	Checkpoint files
yolo12l	`yolo12l/`	`yolo12l/model.onnx` (ONNX) · `yolo12l/best_model_epoch_9.pth` (PyTorch)
yolo12m	`yolo12m/`	`yolo12m/model.onnx` (ONNX) · `yolo12m/best_model_epoch_9.pth` (PyTorch)
yolo12s	`yolo12s/`	`yolo12s/model.onnx` (ONNX) · `yolo12s/best_model_epoch_6.pth` (PyTorch)
yolo12n	`yolo12n/`	`yolo12n/model.onnx` (ONNX) · `yolo12n/best_model_epoch_5.pth` (PyTorch)
yolov8m	`yolov8m/`	`yolov8m/model.onnx` (ONNX) · `yolov8m/best_model_epoch_6.pth` (PyTorch)

Usage

ONNX (recommended — no Python dependencies beyond onnxruntime)

python

from huggingface_hub import hf_hub_download

onnx_path = hf_hub_download(
    repo_id="maelic/REACTPlusPlus_PSG",
    filename="yolo12l/react_pp_yolo12m.onnx",
    repo_type="model",
)
# Run with tools/eval_onnx_psg.py or load directly via onnxruntime

PyTorch

python

# 1. Clone the repository
#    git clone https://github.com/Maelic/SGG-Benchmark

# 2. Install dependencies
#    pip install -e .

# 3. Download checkpoint + config
from huggingface_hub import hf_hub_download

ckpt_path = hf_hub_download(
    repo_id="maelic/REACTPlusPlus_PSG",
    filename="yolo12l/best_model.pth",
    repo_type="model",
)
cfg_path = hf_hub_download(
    repo_id="maelic/REACTPlusPlus_PSG",
    filename="yolo12l/config.yml",
    repo_type="model",
)

# 4. Run evaluation
import subprocess
subprocess.run([
    "python", "tools/relation_eval_hydra.py",
    "--config-path", str(cfg_path),
    "--task", "sgdet",
    "--eval-only",
    "--checkpoint", str(ckpt_path),
])

Citation

bibtex

@article{neau2026reactpp,
  title   = {REACT++: Efficient Cross-Attention for Real-Time Scene Graph Generation
},
  author  = {Neau, Maëlic and Falomir, Zoe},
  year    = {2026},
  url     = {https://arxiv.org/abs/2603.06386},
}