Hugging Face's logo

maelic
/
REACTPlusPlus_VG150

Object Detection
ONNX
PyTorch
English
sgg-benchmark
scene-graph-generation
visual-relationship-detection
yolo
Eval Results (legacy)
Model card Files
xet
 Community
maelic commited on
Commit
2b81988
·
verified ·
1 Parent(s): 332c90a

Update model card

Browse files
Files changed (1) hide show
  1. README.md +32 -14
README.md CHANGED
@@ -38,9 +38,15 @@ model-index:
38
  - type: R@100
39
  value: 28.47
40
  name: R@100
41
- - type: mean_recall
42
- value: 22.54
43
- name: mean_recall
 
 
 
 
 
 
44
  - name: REACT++ yolo26m
45
  results:
46
  - task:
@@ -68,9 +74,15 @@ model-index:
68
  - type: R@100
69
  value: 33.7
70
  name: R@100
71
- - type: mean_recall
72
- value: 21.87
73
- name: mean_recall
 
 
 
 
 
 
74
  - name: REACT++ yolov8m
75
  results:
76
  - task:
@@ -98,9 +110,15 @@ model-index:
98
  - type: R@100
99
  value: 34.09
100
  name: R@100
101
- - type: mean_recall
102
- value: 23.02
103
- name: mean_recall
 
 
 
 
 
 
104
  ---
105
 
106
  # REACT++ Scene Graph Generation — VG150 (yolo12m, yolo26m, yolov8m)
@@ -126,13 +144,13 @@ The models were trained with the
126
 
127
  ## Results — SGDet on VG150 test split (CUDA, max_det=100, batch_size=1)
128
 
129
- > Metrics from end-to-end evaluation (`tools/evaluate.py`). E2E Latency = pre-process + model forward.
130
 
131
- | Backbone | Params | R@20 | R@50 | R@100 | mR@20 | mR@50 | mR@100 | F1@20 | F1@50 | F1@100 | E2E Lat. (ms) |
132
  |----------|:------:|-----:|-----:|------:|------:|------:|-------:|------:|------:|-------:|--------------:|
133
- | yolo12m | ~20.2M | 18.76 | 24.63 | 28.47 | 10.81 | 14.42 | 16.78 | - | - | - | - |
134
- | yolo26m | ~20.2M | 21.12 | 28.34 | 33.7 | 10.81 | 14.6 | 18.36 | - | - | - | - |
135
- | yolov8m | ~25.9M | 22.89 | 29.96 | 34.09 | 12.22 | 16.31 | 18.45 | - | - | - | - |
136
 
137
  ---
138
 
 
38
  - type: R@100
39
  value: 28.47
40
  name: R@100
41
+ - type: F1@20
42
+ value: 13.72
43
+ name: F1@20
44
+ - type: F1@50
45
+ value: 18.19
46
+ name: F1@50
47
+ - type: F1@100
48
+ value: 21.11
49
+ name: F1@100
50
  - name: REACT++ yolo26m
51
  results:
52
  - task:
 
74
  - type: R@100
75
  value: 33.7
76
  name: R@100
77
+ - type: F1@20
78
+ value: 14.3
79
+ name: F1@20
80
+ - type: F1@50
81
+ value: 19.27
82
+ name: F1@50
83
+ - type: F1@100
84
+ value: 23.77
85
+ name: F1@100
86
  - name: REACT++ yolov8m
87
  results:
88
  - task:
 
110
  - type: R@100
111
  value: 34.09
112
  name: R@100
113
+ - type: F1@20
114
+ value: 15.93
115
+ name: F1@20
116
+ - type: F1@50
117
+ value: 21.12
118
+ name: F1@50
119
+ - type: F1@100
120
+ value: 23.94
121
+ name: F1@100
122
  ---
123
 
124
  # REACT++ Scene Graph Generation — VG150 (yolo12m, yolo26m, yolov8m)
 
144
 
145
  ## Results — SGDet on VG150 test split (CUDA, max_det=100, batch_size=1)
146
 
147
+ > Metrics from end-to-end evaluation (`tools/evaluate.py`). Latency = model forward only.
148
 
149
+ | Backbone | Params | R@20 | R@50 | R@100 | mR@20 | mR@50 | mR@100 | F1@20 | F1@50 | F1@100 | Lat. (ms) |
150
  |----------|:------:|-----:|-----:|------:|------:|------:|-------:|------:|------:|-------:|--------------:|
151
+ | yolo12m | ~20.2M | 18.76 | 24.63 | 28.47 | 10.81 | 14.42 | 16.78 | 13.72 | 18.19 | 21.11 | - |
152
+ | yolo26m | ~20.2M | 21.12 | 28.34 | 33.7 | 10.81 | 14.6 | 18.36 | 14.3 | 19.27 | 23.77 | - |
153
+ | yolov8m | ~25.9M | 22.89 | 29.96 | 34.09 | 12.22 | 16.31 | 18.45 | 15.93 | 21.12 | 23.94 | - |
154
 
155
  ---
156