Update model card

README.md

@@ -19,7 +19,37 @@ model-index:
         dataset:
           name: PSG
           type: psg
-        metrics: []
+        metrics:
+          - type: mR@20
+            value: 16.88
+            name: mR@20
+          - type: R@20
+            value: 26.88
+            name: R@20
+          - type: F1@20
+            value: 20.74
+            name: F1@20
+          - type: mR@50
+            value: 18.65
+            name: mR@50
+          - type: R@50
+            value: 30.61
+            name: R@50
+          - type: F1@50
+            value: 23.17
+            name: F1@50
+          - type: mR@100
+            value: 19.5
+            name: mR@100
+          - type: R@100
+            value: 31.8
+            name: R@100
+          - type: F1@100
+            value: 24.17
+            name: F1@100
+          - type: e2e_latency_ms
+            value: 11.4
+            name: e2e_latency_ms
   - name: REACT++ yolo12s
     results:
       - task:
@@ -30,35 +60,35 @@ model-index:
           type: psg
         metrics:
           - type: mR@20
-            value: 2.91
+            value: 21.12
             name: mR@20
           - type: R@20
-            value: 6.71
+            value: 29.28
             name: R@20
-          - type: zsR@20
-            value: 1.82
-            name: zsR@20
+          - type: F1@20
+            value: 24.54
+            name: F1@20
           - type: mR@50
-            value: 3.93
+            value: 23.21
             name: mR@50
           - type: R@50
-            value: 9.28
+            value: 33.48
             name: R@50
-          - type: zsR@50
-            value: 2.66
-            name: zsR@50
+          - type: F1@50
+            value: 27.41
+            name: F1@50
           - type: mR@100
-            value: 4.62
+            value: 23.77
             name: mR@100
           - type: R@100
-            value: 11.21
+            value: 34.74
             name: R@100
-          - type: zsR@100
-            value: 3.22
-            name: zsR@100
-          - type: mean_recall
-            value: 24.71
-            name: mean_recall
+          - type: F1@100
+            value: 28.23
+            name: F1@100
+          - type: e2e_latency_ms
+            value: 12.2
+            name: e2e_latency_ms
   - name: REACT++ yolo12m
     results:
       - task:
@@ -69,35 +99,35 @@ model-index:
           type: psg
         metrics:
           - type: mR@20
-            value: 22.73
+            value: 22.74
             name: mR@20
           - type: R@20
-            value: 31.11
+            value: 32.69
             name: R@20
-          - type: zsR@20
-            value: 1.81
-            name: zsR@20
+          - type: F1@20
+            value: 26.82
+            name: F1@20
           - type: mR@50
-            value: 25.75
+            value: 25.21
             name: mR@50
           - type: R@50
-            value: 36.29
+            value: 37.2
             name: R@50
-          - type: zsR@50
-            value: 2.8
-            name: zsR@50
+          - type: F1@50
+            value: 30.05
+            name: F1@50
           - type: mR@100
-            value: 27.55
+            value: 26.08
             name: mR@100
           - type: R@100
-            value: 39.44
+            value: 38.58
             name: R@100
-          - type: zsR@100
-            value: 3.77
-            name: zsR@100
-          - type: mean_recall
-            value: 26.32
-            name: mean_recall
+          - type: F1@100
+            value: 31.12
+            name: F1@100
+          - type: e2e_latency_ms
+            value: 15.7
+            name: e2e_latency_ms
   - name: REACT++ yolo12l
     results:
       - task:
@@ -108,35 +138,35 @@ model-index:
           type: psg
         metrics:
           - type: mR@20
-            value: 23.34
+            value: 23.2
             name: mR@20
           - type: R@20
-            value: 29.72
+            value: 30.99
             name: R@20
-          - type: zsR@20
-            value: 1.74
-            name: zsR@20
+          - type: F1@20
+            value: 26.53
+            name: F1@20
           - type: mR@50
-            value: 25.82
+            value: 25.49
             name: mR@50
           - type: R@50
-            value: 35.12
+            value: 35.3
             name: R@50
-          - type: zsR@50
-            value: 2.77
-            name: zsR@50
+          - type: F1@50
+            value: 29.6
+            name: F1@50
           - type: mR@100
-            value: 27.47
+            value: 26.45
             name: mR@100
           - type: R@100
-            value: 37.99
+            value: 36.68
             name: R@100
-          - type: zsR@100
-            value: 3.53
-            name: zsR@100
-          - type: mean_recall
-            value: 33.16
-            name: mean_recall
+          - type: F1@100
+            value: 30.74
+            name: F1@100
+          - type: e2e_latency_ms
+            value: 19.6
+            name: e2e_latency_ms
   - name: REACT++ yolov8m
     results:
       - task:
@@ -147,35 +177,35 @@ model-index:
           type: psg
         metrics:
           - type: mR@20
-            value: 2.82
+            value: 22.75
             name: mR@20
           - type: R@20
-            value: 10.02
+            value: 30.69
             name: R@20
-          - type: zsR@20
-            value: 1.97
-            name: zsR@20
+          - type: F1@20
+            value: 26.13
+            name: F1@20
           - type: mR@50
-            value: 4.57
+            value: 25.46
             name: mR@50
           - type: R@50
-            value: 13.75
+            value: 35.68
             name: R@50
-          - type: zsR@50
-            value: 2.8
-            name: zsR@50
+          - type: F1@50
+            value: 29.72
+            name: F1@50
           - type: mR@100
-            value: 5.98
+            value: 26.4
             name: mR@100
           - type: R@100
-            value: 16.24
+            value: 37.43
             name: R@100
-          - type: zsR@100
-            value: 3.49
-            name: zsR@100
-          - type: mean_recall
-            value: 21.42
-            name: mean_recall
+          - type: F1@100
+            value: 30.96
+            name: F1@100
+          - type: e2e_latency_ms
+            value: 15.3
+            name: e2e_latency_ms
 ---
 
 # REACT++ Scene Graph Generation — PSG (yolo12n, yolo12s, yolo12m, yolo12l, yolov8m)
@@ -186,44 +216,62 @@ on the **PSG** benchmark, across 5 backbone sizes.
 REACT++ is a parameter-efficient, attention-augmented relation predictor built on top of
 a YOLO12 backbone.  It uses:
 
+- **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 × Semantic cross-attention** — visual tokens attend to GloVe prototype embeddings
+- **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 (MoCo/DINO-style)
+- **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](https://github.com/Maelic/SGG-Benchmark) framework and described in the
-[REACT paper (Neau et al., BMVC 2025)](https://arxiv.org/abs/2405.16116).
+[REACT++ paper (Neau et al., 2026)](https://arxiv.org/abs/2603.06386).
 
 ---
 
-## Results — SGDet on PSG test split
+## Results — SGDet on PSG test split (ONNX, CUDA)
 
-| Backbone | Params (backbone) | mR@20 | mR@50 | mR@100 | R@20 | R@50 | R@100 |
-|----------|:-----------------:|------:|------:|-------:|-----:|-----:|------:|
-| yolo12n | ~2.6M | - | - | - | - | - | - |
-| yolo12s | ~9.2M | 2.91 | 3.93 | 4.62 | 6.71 | 9.28 | 11.21 |
-| yolo12m | ~20.2M | 22.73 | 25.75 | 27.55 | 31.11 | 36.29 | 39.44 |
-| yolo12l | ~26.5M | 23.34 | 25.82 | 27.47 | 29.72 | 35.12 | 37.99 |
-| yolov8m | ~25.9M | 2.82 | 4.57 | 5.98 | 10.02 | 13.75 | 16.24 |
+> 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) |
+|----------|:------:|-----:|-----:|------:|------:|------:|-------:|------:|------:|-------:|--------------:|
+| yolo12n | ~2.6M | 26.88 | 30.61 | 31.8 | 16.88 | 18.65 | 19.5 | 20.74 | 23.17 | 24.17 | 11.4 |
+| yolo12s | ~9.2M | 29.28 | 33.48 | 34.74 | 21.12 | 23.21 | 23.77 | 24.54 | 27.41 | 28.23 | 12.2 |
+| yolo12m | ~20.2M | 32.69 | 37.2 | 38.58 | 22.74 | 25.21 | 26.08 | 26.82 | 30.05 | 31.12 | 15.7 |
+| yolo12l | ~26.5M | 30.99 | 35.3 | 36.68 | 23.2 | 25.49 | 26.45 | 26.53 | 29.6 | 30.74 | 19.6 |
+| 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 file |
+| Variant | Sub-folder | Checkpoint files |
 |---------|------------|-----------------|
-| yolo12n | `yolo12n/` | `yolo12n/best_model_epoch_5.pth` |
-| yolo12s | `yolo12s/` | `yolo12s/best_model_epoch_6.pth` |
-| yolo12m | `yolo12m/` | `yolo12m/best_model_epoch_9.pth` |
-| yolo12l | `yolo12l/` | `yolo12l/best_model_epoch_9.pth` |
-| yolov8m | `yolov8m/` | `yolov8m/best_model_epoch_6.pth` |
+| yolo12n | `yolo12n/` | `yolo12n/model.onnx` (ONNX) · `yolo12n/best_model_epoch_5.pth` (PyTorch) |
+| yolo12s | `yolo12s/` | `yolo12s/model.onnx` (ONNX) · `yolo12s/best_model_epoch_6.pth` (PyTorch) |
+| yolo12m | `yolo12m/` | `yolo12m/model.onnx` (ONNX) · `yolo12m/best_model_epoch_9.pth` (PyTorch) |
+| yolo12l | `yolo12l/` | `yolo12l/model.onnx` (ONNX) · `yolo12l/best_model_epoch_9.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="yolo12n/model.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
@@ -231,7 +279,7 @@ The models were trained with the
 # 2. Install dependencies
 #    pip install -e .
 
-# 3. Download a checkpoint
+# 3. Download checkpoint + config
 from huggingface_hub import hf_hub_download
 
 ckpt_path = hf_hub_download(
@@ -248,7 +296,7 @@ cfg_path = hf_hub_download(
 # 4. Run evaluation
 import subprocess
 subprocess.run([
-    "python", "tools/relation_train_net_hydra.py",
+    "python", "tools/relation_eval_hydra.py",
     "--config-path", str(cfg_path),
     "--task", "sgdet",
     "--eval-only",
@@ -261,11 +309,11 @@ subprocess.run([
 ## Citation
 
 ```bibtex
-@inproceedings{neau2025react,
-  title   = {REACT: Relation Extraction through Attention-guided Contrastive Training},
-  author  = {Neau, Maëlic and others},
-  booktitle = {BMVC},
-  year    = {2025},
-  url     = {https://arxiv.org/abs/2405.16116},
+@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},
 }
 ```