sebastientaylor commited on
Commit
b8499fe
Β·
verified Β·
1 Parent(s): 7787cf3

Update model card for yolov8-det

Browse files
Files changed (1) hide show
  1. README.md +22 -14
README.md CHANGED
@@ -28,7 +28,7 @@ model-index:
28
  value: 50.55
29
  - name: "mAP@0.5-0.95 (Nano ONNX FP32)"
30
  type: map
31
- value: 35.87
32
  - name: "mAP@0.5 (Small ONNX FP32)"
33
  type: map_50
34
  value: 59.39
@@ -37,10 +37,10 @@ model-index:
37
  value: 43.23
38
  - name: "mAP@0.5 (Medium ONNX FP32)"
39
  type: map_50
40
- value: 64.7
41
  - name: "mAP@0.5-0.95 (Medium ONNX FP32)"
42
  type: map
43
- value: 48.46
44
  ---
45
 
46
  # YOLOv8 Detection β€” EdgeFirst Model Zoo
@@ -63,9 +63,9 @@ Accuracy ceiling for each size, measured against COCO `val2017` (5,000 images) w
63
 
64
  | Size | Params | GFLOPs | mAP@0.5 | mAP@0.5-0.95 | mAP@0.75 | Source |
65
  |------|--------|--------|---------|--------------|----------|--------|
66
- | Nano | 3.2M | 8.9 | 50.55% | 35.87% | 38.92% | [v-1d30](https://test.edgefirst.studio/public/validation/v-1d30/details?mode=info) |
67
- | Small | 11.2M | 28.8 | 59.39% | 43.23% | 46.69% | [v-1d33](https://test.edgefirst.studio/public/validation/v-1d33/details?mode=info) |
68
- | Medium | 25.9M | 79.3 | 64.70% | 48.46% | 52.62% | [v-1d34](https://test.edgefirst.studio/public/validation/v-1d34/details?mode=info) |
69
  | Large | 43.7M | 165.7 | β€” | β€” | β€” | β€” |
70
  | XLarge | 68.2M | 258.5 | β€” | β€” | β€” | β€” |
71
 
@@ -79,13 +79,21 @@ Rows without a number under the metric columns are validation sessions that are
79
 
80
  | Size | Platform | mAP@0.5 | Ξ” vs FP32 (pp) | mAP@0.5-0.95 | Inference (ms) | FPS (median) | Source |
81
  |------|----------|---------|----------------|--------------|----------------|--------------|--------|
82
- | Nano | NXP i.MX 8M Plus + VeriSilicon NPU | 48.71% | -1.84 | 34.19% | 61.75 | 12.1 | [v-1d45](https://test.edgefirst.studio/public/validation/v-1d45/details?mode=info) |
83
- | Nano | NXP i.MX 95 + eIQ Neutron NPU | 48.82% | -1.73 | 34.28% | 13.63 | 53.6 | [v-1d42](https://test.edgefirst.studio/public/validation/v-1d42/details?mode=info) |
84
- | Nano | Raspberry Pi 5 + Hailo-8L NPU | 49.00% | -1.55 | 34.62% | 15.01 | 63.9 | [v-1d29](https://test.edgefirst.studio/public/validation/v-1d29/details?mode=info) |
85
- | Small | NXP i.MX 95 + eIQ Neutron NPU | 58.67% | -0.72 | 42.41% | 36.90 | 24.0 | [v-1d46](https://test.edgefirst.studio/public/validation/v-1d46/details?mode=info) |
86
- | Small | Raspberry Pi 5 + Hailo-8L NPU | 58.36% | -1.03 | 42.18% | 33.98 | 28.8 | [v-1d2a](https://test.edgefirst.studio/public/validation/v-1d2a/details?mode=info) |
87
- | Medium | NXP i.MX 95 + eIQ Neutron NPU | 63.97% | -0.73 | 47.71% | 99.94 | 9.4 | [v-1d4a](https://test.edgefirst.studio/public/validation/v-1d4a/details?mode=info) |
88
- | Medium | Raspberry Pi 5 + Hailo-8L NPU | 64.02% | -0.68 | 47.79% | 58.20 | 17.0 | [v-1d31](https://test.edgefirst.studio/public/validation/v-1d31/details?mode=info) |
 
 
 
 
 
 
 
 
89
 
90
 
91
  ---
@@ -138,7 +146,7 @@ for det in results.detections:
138
  Every measurement in the tables above is reachable through the EdgeFirst Studio validation framework. The `v-XXXX` Source link on each row resolves to a public Studio URL of the form:
139
 
140
  ```
141
- https://test.edgefirst.studio/public/validation/v-XXXX/details?mode=info
142
  ```
143
 
144
  From there, the full provenance chain is one click deeper: training session ID, dataset version, calibration artifact, converter chain (e.g. TFLite quantizer + Neutron compile), validation parameters, and the host hardware description (hostname, kernel version, SoC, NPU, profiler version). The same model file you download from this repository embeds the same chain in its `edgefirst.json` metadata.
 
28
  value: 50.55
29
  - name: "mAP@0.5-0.95 (Nano ONNX FP32)"
30
  type: map
31
+ value: 35.82
32
  - name: "mAP@0.5 (Small ONNX FP32)"
33
  type: map_50
34
  value: 59.39
 
37
  value: 43.23
38
  - name: "mAP@0.5 (Medium ONNX FP32)"
39
  type: map_50
40
+ value: 64.73
41
  - name: "mAP@0.5-0.95 (Medium ONNX FP32)"
42
  type: map
43
+ value: 48.41
44
  ---
45
 
46
  # YOLOv8 Detection β€” EdgeFirst Model Zoo
 
63
 
64
  | Size | Params | GFLOPs | mAP@0.5 | mAP@0.5-0.95 | mAP@0.75 | Source |
65
  |------|--------|--------|---------|--------------|----------|--------|
66
+ | Nano | 3.2M | 8.9 | 50.55% | 35.82% | 38.89% | [v-1db9](https://test.edgefirst.studio/public/validation/v-1db9/details) |
67
+ | Small | 11.2M | 28.8 | 59.39% | 43.23% | 46.69% | [v-1d33](https://test.edgefirst.studio/public/validation/v-1d33/details) |
68
+ | Medium | 25.9M | 79.3 | 64.73% | 48.41% | 52.55% | [v-1dbd](https://test.edgefirst.studio/public/validation/v-1dbd/details) |
69
  | Large | 43.7M | 165.7 | β€” | β€” | β€” | β€” |
70
  | XLarge | 68.2M | 258.5 | β€” | β€” | β€” | β€” |
71
 
 
79
 
80
  | Size | Platform | mAP@0.5 | Ξ” vs FP32 (pp) | mAP@0.5-0.95 | Inference (ms) | FPS (median) | Source |
81
  |------|----------|---------|----------------|--------------|----------------|--------------|--------|
82
+ | Nano | ONNX FP32 (CPU) | 50.55% | +0.00 | 35.87% | 5.47 | 163.6 | [v-1d30](https://test.edgefirst.studio/public/validation/v-1d30/details) |
83
+ | Nano | ONNX FP32 | 50.55% | +0.00 | 35.84% | 3.07 | 321.6 | [v-1db8](https://test.edgefirst.studio/public/validation/v-1db8/details) |
84
+ | Nano | NXP i.MX 8M Plus + VeriSilicon NPU | 48.71% | -1.84 | 34.19% | 61.75 | 12.1 | [v-1d45](https://test.edgefirst.studio/public/validation/v-1d45/details) |
85
+ | Nano | NXP i.MX 95 + eIQ Neutron NPU | 48.82% | -1.73 | 34.28% | 13.63 | 53.6 | [v-1d42](https://test.edgefirst.studio/public/validation/v-1d42/details) |
86
+ | Nano | Raspberry Pi 5 + Hailo-8L NPU | 49.00% | -1.55 | 34.62% | 15.01 | 63.9 | [v-1d29](https://test.edgefirst.studio/public/validation/v-1d29/details) |
87
+ | Small | ONNX FP32 | 59.35% | -0.04 | 43.18% | 11.21 | 88.7 | [v-1db7](https://test.edgefirst.studio/public/validation/v-1db7/details) |
88
+ | Small | ONNX FP32 | 59.36% | -0.03 | 43.21% | 6.15 | 162.7 | [v-1db6](https://test.edgefirst.studio/public/validation/v-1db6/details) |
89
+ | Small | NXP i.MX 95 + eIQ Neutron NPU | 58.67% | -0.72 | 42.41% | 36.90 | 24.0 | [v-1d46](https://test.edgefirst.studio/public/validation/v-1d46/details) |
90
+ | Small | Raspberry Pi 5 + Hailo-8L NPU | 58.36% | -1.03 | 42.18% | 33.98 | 28.8 | [v-1d2a](https://test.edgefirst.studio/public/validation/v-1d2a/details) |
91
+ | Medium | ONNX FP32 (CPU) | 64.70% | -0.03 | 48.46% | 9.14 | 105.2 | [v-1d34](https://test.edgefirst.studio/public/validation/v-1d34/details) |
92
+ | Medium | ONNX FP32 | 64.68% | -0.05 | 48.46% | 14.10 | 70.7 | [v-1dbc](https://test.edgefirst.studio/public/validation/v-1dbc/details) |
93
+ | Medium | ONNX FP32 | 64.73% | +0.00 | 48.41% | 13.82 | 72.3 | [v-1db5](https://test.edgefirst.studio/public/validation/v-1db5/details) |
94
+ | Medium | ONNX FP32 | 64.68% | -0.05 | 48.46% | 14.16 | 70.6 | [v-1db4](https://test.edgefirst.studio/public/validation/v-1db4/details) |
95
+ | Medium | NXP i.MX 95 + eIQ Neutron NPU | 63.97% | -0.76 | 47.71% | 99.94 | 9.4 | [v-1d4a](https://test.edgefirst.studio/public/validation/v-1d4a/details) |
96
+ | Medium | Raspberry Pi 5 + Hailo-8L NPU | 64.02% | -0.71 | 47.79% | 58.20 | 17.0 | [v-1d31](https://test.edgefirst.studio/public/validation/v-1d31/details) |
97
 
98
 
99
  ---
 
146
  Every measurement in the tables above is reachable through the EdgeFirst Studio validation framework. The `v-XXXX` Source link on each row resolves to a public Studio URL of the form:
147
 
148
  ```
149
+ https://test.edgefirst.studio/public/validation/v-XXXX/details
150
  ```
151
 
152
  From there, the full provenance chain is one click deeper: training session ID, dataset version, calibration artifact, converter chain (e.g. TFLite quantizer + Neutron compile), validation parameters, and the host hardware description (hostname, kernel version, SoC, NPU, profiler version). The same model file you download from this repository embeds the same chain in its `edgefirst.json` metadata.