qaihm-bot commited on
Commit
e0e561f
·
verified ·
1 Parent(s): 8d5e7cf

See https://github.com/quic/ai-hub-models/releases/v0.34.0 for changelog.

Files changed (1) hide show
  1. README.md +241 -88
README.md CHANGED
@@ -19,7 +19,11 @@ Ultralytics YOLOv8 is a machine learning model that predicts bounding boxes and
19
  This model is an implementation of YOLOv8-Detection found [here](https://github.com/ultralytics/ultralytics/tree/main/ultralytics/models/yolo/detect).
20
 
21
 
22
- More details on model performance across various devices, can be found [here](https://aihub.qualcomm.com/models/yolov8_det).
 
 
 
 
23
 
24
  ### Model Details
25
 
@@ -34,87 +38,255 @@ This model is an implementation of YOLOv8-Detection found [here](https://github.
34
 
35
  | Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model
36
  |---|---|---|---|---|---|---|---|---|
37
- | YOLOv8-Detection | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 16.523 ms | 0 - 40 MB | NPU | -- |
38
  | YOLOv8-Detection | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 13.669 ms | 3 - 98 MB | NPU | -- |
39
- | YOLOv8-Detection | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 8.449 ms | 0 - 38 MB | NPU | -- |
40
  | YOLOv8-Detection | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 8.269 ms | 5 - 44 MB | NPU | -- |
41
- | YOLOv8-Detection | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 5.639 ms | 0 - 22 MB | NPU | -- |
42
  | YOLOv8-Detection | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 3.551 ms | 0 - 54 MB | NPU | -- |
43
- | YOLOv8-Detection | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 7.189 ms | 0 - 40 MB | NPU | -- |
44
  | YOLOv8-Detection | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 5.212 ms | 2 - 120 MB | NPU | -- |
45
- | YOLOv8-Detection | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 16.523 ms | 0 - 40 MB | NPU | -- |
46
  | YOLOv8-Detection | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 13.669 ms | 3 - 98 MB | NPU | -- |
47
- | YOLOv8-Detection | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 5.857 ms | 0 - 20 MB | NPU | -- |
48
  | YOLOv8-Detection | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 3.551 ms | 0 - 55 MB | NPU | -- |
49
- | YOLOv8-Detection | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 9.577 ms | 0 - 27 MB | NPU | -- |
50
  | YOLOv8-Detection | float | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 7.248 ms | 3 - 34 MB | NPU | -- |
51
- | YOLOv8-Detection | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 5.728 ms | 0 - 21 MB | NPU | -- |
52
  | YOLOv8-Detection | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 3.543 ms | 0 - 67 MB | NPU | -- |
53
- | YOLOv8-Detection | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 7.189 ms | 0 - 40 MB | NPU | -- |
54
  | YOLOv8-Detection | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 5.212 ms | 2 - 120 MB | NPU | -- |
55
- | YOLOv8-Detection | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 5.766 ms | 0 - 21 MB | NPU | -- |
56
  | YOLOv8-Detection | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 3.544 ms | 0 - 61 MB | NPU | -- |
57
- | YOLOv8-Detection | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 5.435 ms | 0 - 52 MB | NPU | -- |
58
- | YOLOv8-Detection | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 4.131 ms | 0 - 52 MB | NPU | -- |
59
  | YOLOv8-Detection | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 2.659 ms | 5 - 227 MB | NPU | -- |
60
- | YOLOv8-Detection | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 3.774 ms | 1 - 181 MB | NPU | -- |
61
- | YOLOv8-Detection | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 4.038 ms | 0 - 48 MB | NPU | -- |
62
  | YOLOv8-Detection | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 2.429 ms | 5 - 120 MB | NPU | -- |
63
- | YOLOv8-Detection | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 3.759 ms | 5 - 92 MB | NPU | -- |
64
  | YOLOv8-Detection | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 4.22 ms | 5 - 5 MB | NPU | -- |
65
- | YOLOv8-Detection | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 5.937 ms | 5 - 5 MB | NPU | -- |
66
- | YOLOv8-Detection | w8a16 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 6.927 ms | 1 - 31 MB | NPU | -- |
67
- | YOLOv8-Detection | w8a16 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 4.142 ms | 2 - 43 MB | NPU | -- |
68
- | YOLOv8-Detection | w8a16 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 3.505 ms | 2 - 13 MB | NPU | -- |
69
- | YOLOv8-Detection | w8a16 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 4.038 ms | 2 - 32 MB | NPU | -- |
70
- | YOLOv8-Detection | w8a16 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | QNN_DLC | 13.722 ms | 2 - 37 MB | NPU | -- |
71
- | YOLOv8-Detection | w8a16 | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 6.927 ms | 1 - 31 MB | NPU | -- |
72
- | YOLOv8-Detection | w8a16 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 3.511 ms | 2 - 12 MB | NPU | -- |
73
- | YOLOv8-Detection | w8a16 | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 4.766 ms | 2 - 34 MB | NPU | -- |
74
- | YOLOv8-Detection | w8a16 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 3.471 ms | 2 - 13 MB | NPU | -- |
75
- | YOLOv8-Detection | w8a16 | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 4.038 ms | 2 - 32 MB | NPU | -- |
76
- | YOLOv8-Detection | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 3.5 ms | 2 - 13 MB | NPU | -- |
77
- | YOLOv8-Detection | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 9.309 ms | 3 - 23 MB | NPU | -- |
78
- | YOLOv8-Detection | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 2.336 ms | 2 - 43 MB | NPU | -- |
79
- | YOLOv8-Detection | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 6.749 ms | 2 - 78 MB | NPU | -- |
80
- | YOLOv8-Detection | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 1.993 ms | 2 - 37 MB | NPU | -- |
81
- | YOLOv8-Detection | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 5.965 ms | 2 - 78 MB | NPU | -- |
82
- | YOLOv8-Detection | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 4.116 ms | 9 - 9 MB | NPU | -- |
83
- | YOLOv8-Detection | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 10.049 ms | 1 - 1 MB | NPU | -- |
84
- | YOLOv8-Detection | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 3.304 ms | 0 - 20 MB | NPU | -- |
85
- | YOLOv8-Detection | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 3.323 ms | 1 - 24 MB | NPU | -- |
86
- | YOLOv8-Detection | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 1.642 ms | 0 - 36 MB | NPU | -- |
87
- | YOLOv8-Detection | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 1.757 ms | 1 - 38 MB | NPU | -- |
88
- | YOLOv8-Detection | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 1.519 ms | 0 - 15 MB | NPU | -- |
89
- | YOLOv8-Detection | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 1.517 ms | 1 - 12 MB | NPU | -- |
90
- | YOLOv8-Detection | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 1.89 ms | 0 - 22 MB | NPU | -- |
91
- | YOLOv8-Detection | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 1.927 ms | 1 - 26 MB | NPU | -- |
92
- | YOLOv8-Detection | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | TFLITE | 3.83 ms | 0 - 29 MB | NPU | -- |
93
- | YOLOv8-Detection | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | QNN_DLC | 5.085 ms | 1 - 32 MB | NPU | -- |
94
- | YOLOv8-Detection | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | TFLITE | 49.411 ms | 3 - 19 MB | NPU | -- |
95
- | YOLOv8-Detection | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 3.304 ms | 0 - 20 MB | NPU | -- |
96
- | YOLOv8-Detection | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 3.323 ms | 1 - 24 MB | NPU | -- |
97
- | YOLOv8-Detection | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 1.508 ms | 0 - 14 MB | NPU | -- |
98
- | YOLOv8-Detection | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 1.51 ms | 0 - 10 MB | NPU | -- |
99
- | YOLOv8-Detection | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 2.32 ms | 0 - 27 MB | NPU | -- |
100
- | YOLOv8-Detection | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 2.332 ms | 1 - 27 MB | NPU | -- |
101
- | YOLOv8-Detection | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 1.495 ms | 0 - 15 MB | NPU | -- |
102
- | YOLOv8-Detection | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 1.519 ms | 1 - 12 MB | NPU | -- |
103
- | YOLOv8-Detection | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 1.89 ms | 0 - 22 MB | NPU | -- |
104
- | YOLOv8-Detection | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 1.927 ms | 1 - 26 MB | NPU | -- |
105
- | YOLOv8-Detection | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 1.495 ms | 0 - 15 MB | NPU | -- |
106
- | YOLOv8-Detection | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 1.515 ms | 1 - 12 MB | NPU | -- |
107
- | YOLOv8-Detection | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 5.35 ms | 0 - 26 MB | NPU | -- |
108
- | YOLOv8-Detection | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 0.982 ms | 0 - 36 MB | NPU | -- |
109
- | YOLOv8-Detection | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 1.027 ms | 1 - 39 MB | NPU | -- |
110
- | YOLOv8-Detection | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 3.941 ms | 0 - 89 MB | NPU | -- |
111
- | YOLOv8-Detection | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 0.906 ms | 0 - 25 MB | NPU | -- |
112
- | YOLOv8-Detection | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 0.881 ms | 1 - 32 MB | NPU | -- |
113
- | YOLOv8-Detection | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 2.828 ms | 1 - 78 MB | NPU | -- |
114
- | YOLOv8-Detection | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 1.802 ms | 7 - 7 MB | NPU | -- |
115
- | YOLOv8-Detection | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 5.93 ms | 0 - 0 MB | NPU | -- |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
116
 
 
 
117
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
118
 
119
 
120
  ## License
@@ -131,26 +303,7 @@ This model is an implementation of YOLOv8-Detection found [here](https://github.
131
 
132
 
133
  ## Community
134
- * Join [our AI Hub Slack community](https://qualcomm-ai-hub.slack.com/join/shared_invite/zt-2d5zsmas3-Sj0Q9TzslueCjS31eXG2UA#/shared-invite/email) to collaborate, post questions and learn more about on-device AI.
135
  * For questions or feedback please [reach out to us](mailto:[email protected]).
136
 
137
- ## Usage and Limitations
138
-
139
- Model may not be used for or in connection with any of the following applications:
140
-
141
- - Accessing essential private and public services and benefits;
142
- - Administration of justice and democratic processes;
143
- - Assessing or recognizing the emotional state of a person;
144
- - Biometric and biometrics-based systems, including categorization of persons based on sensitive characteristics;
145
- - Education and vocational training;
146
- - Employment and workers management;
147
- - Exploitation of the vulnerabilities of persons resulting in harmful behavior;
148
- - General purpose social scoring;
149
- - Law enforcement;
150
- - Management and operation of critical infrastructure;
151
- - Migration, asylum and border control management;
152
- - Predictive policing;
153
- - Real-time remote biometric identification in public spaces;
154
- - Recommender systems of social media platforms;
155
- - Scraping of facial images (from the internet or otherwise); and/or
156
- - Subliminal manipulation
 
19
  This model is an implementation of YOLOv8-Detection found [here](https://github.com/ultralytics/ultralytics/tree/main/ultralytics/models/yolo/detect).
20
 
21
 
22
+ This repository provides scripts to run YOLOv8-Detection on Qualcomm® devices.
23
+ More details on model performance across various devices, can be found
24
+ [here](https://aihub.qualcomm.com/models/yolov8_det).
25
+
26
+ **WARNING**: The model assets are not readily available for download due to licensing restrictions.
27
 
28
  ### Model Details
29
 
 
38
 
39
  | Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model
40
  |---|---|---|---|---|---|---|---|---|
41
+ | YOLOv8-Detection | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 16.469 ms | 0 - 40 MB | NPU | -- |
42
  | YOLOv8-Detection | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 13.669 ms | 3 - 98 MB | NPU | -- |
43
+ | YOLOv8-Detection | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 8.41 ms | 0 - 37 MB | NPU | -- |
44
  | YOLOv8-Detection | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 8.269 ms | 5 - 44 MB | NPU | -- |
45
+ | YOLOv8-Detection | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 5.645 ms | 0 - 19 MB | NPU | -- |
46
  | YOLOv8-Detection | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 3.551 ms | 0 - 54 MB | NPU | -- |
47
+ | YOLOv8-Detection | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 7.207 ms | 0 - 41 MB | NPU | -- |
48
  | YOLOv8-Detection | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 5.212 ms | 2 - 120 MB | NPU | -- |
49
+ | YOLOv8-Detection | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 16.469 ms | 0 - 40 MB | NPU | -- |
50
  | YOLOv8-Detection | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 13.669 ms | 3 - 98 MB | NPU | -- |
51
+ | YOLOv8-Detection | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 5.726 ms | 0 - 21 MB | NPU | -- |
52
  | YOLOv8-Detection | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 3.551 ms | 0 - 55 MB | NPU | -- |
53
+ | YOLOv8-Detection | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 9.601 ms | 0 - 26 MB | NPU | -- |
54
  | YOLOv8-Detection | float | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 7.248 ms | 3 - 34 MB | NPU | -- |
55
+ | YOLOv8-Detection | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 5.705 ms | 0 - 20 MB | NPU | -- |
56
  | YOLOv8-Detection | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 3.543 ms | 0 - 67 MB | NPU | -- |
57
+ | YOLOv8-Detection | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 7.207 ms | 0 - 41 MB | NPU | -- |
58
  | YOLOv8-Detection | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 5.212 ms | 2 - 120 MB | NPU | -- |
59
+ | YOLOv8-Detection | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 5.689 ms | 0 - 21 MB | NPU | -- |
60
  | YOLOv8-Detection | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 3.544 ms | 0 - 61 MB | NPU | -- |
61
+ | YOLOv8-Detection | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 5.638 ms | 0 - 56 MB | NPU | -- |
62
+ | YOLOv8-Detection | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 4.136 ms | 0 - 52 MB | NPU | -- |
63
  | YOLOv8-Detection | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 2.659 ms | 5 - 227 MB | NPU | -- |
64
+ | YOLOv8-Detection | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 3.902 ms | 5 - 177 MB | NPU | -- |
65
+ | YOLOv8-Detection | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 4.039 ms | 0 - 49 MB | NPU | -- |
66
  | YOLOv8-Detection | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 2.429 ms | 5 - 120 MB | NPU | -- |
67
+ | YOLOv8-Detection | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 2.992 ms | 4 - 89 MB | NPU | -- |
68
  | YOLOv8-Detection | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 4.22 ms | 5 - 5 MB | NPU | -- |
69
+ | YOLOv8-Detection | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 5.967 ms | 5 - 5 MB | NPU | -- |
70
+ | YOLOv8-Detection | w8a16 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 6.916 ms | 1 - 31 MB | NPU | -- |
71
+ | YOLOv8-Detection | w8a16 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 4.262 ms | 2 - 40 MB | NPU | -- |
72
+ | YOLOv8-Detection | w8a16 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 3.51 ms | 2 - 12 MB | NPU | -- |
73
+ | YOLOv8-Detection | w8a16 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 4.024 ms | 2 - 34 MB | NPU | -- |
74
+ | YOLOv8-Detection | w8a16 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | QNN_DLC | 12.686 ms | 2 - 37 MB | NPU | -- |
75
+ | YOLOv8-Detection | w8a16 | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 6.916 ms | 1 - 31 MB | NPU | -- |
76
+ | YOLOv8-Detection | w8a16 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 3.485 ms | 2 - 12 MB | NPU | -- |
77
+ | YOLOv8-Detection | w8a16 | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 4.731 ms | 0 - 33 MB | NPU | -- |
78
+ | YOLOv8-Detection | w8a16 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 3.496 ms | 2 - 12 MB | NPU | -- |
79
+ | YOLOv8-Detection | w8a16 | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 4.024 ms | 2 - 34 MB | NPU | -- |
80
+ | YOLOv8-Detection | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 3.484 ms | 2 - 12 MB | NPU | -- |
81
+ | YOLOv8-Detection | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 8.746 ms | 0 - 27 MB | NPU | -- |
82
+ | YOLOv8-Detection | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 2.32 ms | 2 - 41 MB | NPU | -- |
83
+ | YOLOv8-Detection | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 6.238 ms | 3 - 78 MB | NPU | -- |
84
+ | YOLOv8-Detection | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 1.962 ms | 2 - 36 MB | NPU | -- |
85
+ | YOLOv8-Detection | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 5.835 ms | 2 - 75 MB | NPU | -- |
86
+ | YOLOv8-Detection | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 4.118 ms | 8 - 8 MB | NPU | -- |
87
+ | YOLOv8-Detection | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 9.776 ms | 2 - 2 MB | NPU | -- |
88
+ | YOLOv8-Detection | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 3.348 ms | 0 - 21 MB | NPU | -- |
89
+ | YOLOv8-Detection | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 3.326 ms | 1 - 24 MB | NPU | -- |
90
+ | YOLOv8-Detection | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 1.668 ms | 0 - 33 MB | NPU | -- |
91
+ | YOLOv8-Detection | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 1.757 ms | 1 - 39 MB | NPU | -- |
92
+ | YOLOv8-Detection | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 1.49 ms | 0 - 15 MB | NPU | -- |
93
+ | YOLOv8-Detection | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 1.494 ms | 1 - 12 MB | NPU | -- |
94
+ | YOLOv8-Detection | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 1.87 ms | 0 - 22 MB | NPU | -- |
95
+ | YOLOv8-Detection | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 1.901 ms | 1 - 25 MB | NPU | -- |
96
+ | YOLOv8-Detection | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | TFLITE | 3.768 ms | 0 - 30 MB | NPU | -- |
97
+ | YOLOv8-Detection | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | QNN_DLC | 5.01 ms | 1 - 30 MB | NPU | -- |
98
+ | YOLOv8-Detection | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | TFLITE | 50.364 ms | 2 - 11 MB | NPU | -- |
99
+ | YOLOv8-Detection | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 3.348 ms | 0 - 21 MB | NPU | -- |
100
+ | YOLOv8-Detection | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 3.326 ms | 1 - 24 MB | NPU | -- |
101
+ | YOLOv8-Detection | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 1.511 ms | 0 - 15 MB | NPU | -- |
102
+ | YOLOv8-Detection | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 1.512 ms | 1 - 11 MB | NPU | -- |
103
+ | YOLOv8-Detection | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 2.295 ms | 0 - 27 MB | NPU | -- |
104
+ | YOLOv8-Detection | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 2.329 ms | 0 - 25 MB | NPU | -- |
105
+ | YOLOv8-Detection | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 1.509 ms | 0 - 15 MB | NPU | -- |
106
+ | YOLOv8-Detection | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 1.514 ms | 1 - 12 MB | NPU | -- |
107
+ | YOLOv8-Detection | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 1.87 ms | 0 - 22 MB | NPU | -- |
108
+ | YOLOv8-Detection | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 1.901 ms | 1 - 25 MB | NPU | -- |
109
+ | YOLOv8-Detection | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 1.504 ms | 0 - 15 MB | NPU | -- |
110
+ | YOLOv8-Detection | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 1.513 ms | 0 - 10 MB | NPU | -- |
111
+ | YOLOv8-Detection | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 5.389 ms | 0 - 24 MB | NPU | -- |
112
+ | YOLOv8-Detection | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 0.99 ms | 0 - 34 MB | NPU | -- |
113
+ | YOLOv8-Detection | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 1.019 ms | 1 - 33 MB | NPU | -- |
114
+ | YOLOv8-Detection | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 3.838 ms | 1 - 79 MB | NPU | -- |
115
+ | YOLOv8-Detection | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 0.907 ms | 0 - 22 MB | NPU | -- |
116
+ | YOLOv8-Detection | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 0.868 ms | 1 - 30 MB | NPU | -- |
117
+ | YOLOv8-Detection | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 3.34 ms | 0 - 79 MB | NPU | -- |
118
+ | YOLOv8-Detection | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 1.808 ms | 7 - 7 MB | NPU | -- |
119
+ | YOLOv8-Detection | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 5.731 ms | 2 - 2 MB | NPU | -- |
120
+
121
+
122
+
123
+
124
+ ## Installation
125
+
126
+
127
+ Install the package via pip:
128
+ ```bash
129
+ pip install "qai-hub-models[yolov8-det]"
130
+ ```
131
+
132
+
133
+ ## Configure Qualcomm® AI Hub to run this model on a cloud-hosted device
134
+
135
+ Sign-in to [Qualcomm® AI Hub](https://app.aihub.qualcomm.com/) with your
136
+ Qualcomm® ID. Once signed in navigate to `Account -> Settings -> API Token`.
137
+
138
+ With this API token, you can configure your client to run models on the cloud
139
+ hosted devices.
140
+ ```bash
141
+ qai-hub configure --api_token API_TOKEN
142
+ ```
143
+ Navigate to [docs](https://app.aihub.qualcomm.com/docs/) for more information.
144
+
145
+
146
+
147
+ ## Demo off target
148
+
149
+ The package contains a simple end-to-end demo that downloads pre-trained
150
+ weights and runs this model on a sample input.
151
+
152
+ ```bash
153
+ python -m qai_hub_models.models.yolov8_det.demo
154
+ ```
155
+
156
+ The above demo runs a reference implementation of pre-processing, model
157
+ inference, and post processing.
158
+
159
+ **NOTE**: If you want running in a Jupyter Notebook or Google Colab like
160
+ environment, please add the following to your cell (instead of the above).
161
+ ```
162
+ %run -m qai_hub_models.models.yolov8_det.demo
163
+ ```
164
+
165
+
166
+ ### Run model on a cloud-hosted device
167
+
168
+ In addition to the demo, you can also run the model on a cloud-hosted Qualcomm®
169
+ device. This script does the following:
170
+ * Performance check on-device on a cloud-hosted device
171
+ * Downloads compiled assets that can be deployed on-device for Android.
172
+ * Accuracy check between PyTorch and on-device outputs.
173
+
174
+ ```bash
175
+ python -m qai_hub_models.models.yolov8_det.export
176
+ ```
177
+
178
+
179
+
180
+ ## How does this work?
181
+
182
+ This [export script](https://aihub.qualcomm.com/models/yolov8_det/qai_hub_models/models/YOLOv8-Detection/export.py)
183
+ leverages [Qualcomm® AI Hub](https://aihub.qualcomm.com/) to optimize, validate, and deploy this model
184
+ on-device. Lets go through each step below in detail:
185
+
186
+ Step 1: **Compile model for on-device deployment**
187
+
188
+ To compile a PyTorch model for on-device deployment, we first trace the model
189
+ in memory using the `jit.trace` and then call the `submit_compile_job` API.
190
+
191
+ ```python
192
+ import torch
193
 
194
+ import qai_hub as hub
195
+ from qai_hub_models.models.yolov8_det import Model
196
 
197
+ # Load the model
198
+ torch_model = Model.from_pretrained()
199
+
200
+ # Device
201
+ device = hub.Device("Samsung Galaxy S24")
202
+
203
+ # Trace model
204
+ input_shape = torch_model.get_input_spec()
205
+ sample_inputs = torch_model.sample_inputs()
206
+
207
+ pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])
208
+
209
+ # Compile model on a specific device
210
+ compile_job = hub.submit_compile_job(
211
+ model=pt_model,
212
+ device=device,
213
+ input_specs=torch_model.get_input_spec(),
214
+ )
215
+
216
+ # Get target model to run on-device
217
+ target_model = compile_job.get_target_model()
218
+
219
+ ```
220
+
221
+
222
+ Step 2: **Performance profiling on cloud-hosted device**
223
+
224
+ After compiling models from step 1. Models can be profiled model on-device using the
225
+ `target_model`. Note that this scripts runs the model on a device automatically
226
+ provisioned in the cloud. Once the job is submitted, you can navigate to a
227
+ provided job URL to view a variety of on-device performance metrics.
228
+ ```python
229
+ profile_job = hub.submit_profile_job(
230
+ model=target_model,
231
+ device=device,
232
+ )
233
+
234
+ ```
235
+
236
+ Step 3: **Verify on-device accuracy**
237
+
238
+ To verify the accuracy of the model on-device, you can run on-device inference
239
+ on sample input data on the same cloud hosted device.
240
+ ```python
241
+ input_data = torch_model.sample_inputs()
242
+ inference_job = hub.submit_inference_job(
243
+ model=target_model,
244
+ device=device,
245
+ inputs=input_data,
246
+ )
247
+ on_device_output = inference_job.download_output_data()
248
+
249
+ ```
250
+ With the output of the model, you can compute like PSNR, relative errors or
251
+ spot check the output with expected output.
252
+
253
+ **Note**: This on-device profiling and inference requires access to Qualcomm®
254
+ AI Hub. [Sign up for access](https://myaccount.qualcomm.com/signup).
255
+
256
+
257
+
258
+ ## Run demo on a cloud-hosted device
259
+
260
+ You can also run the demo on-device.
261
+
262
+ ```bash
263
+ python -m qai_hub_models.models.yolov8_det.demo --eval-mode on-device
264
+ ```
265
+
266
+ **NOTE**: If you want running in a Jupyter Notebook or Google Colab like
267
+ environment, please add the following to your cell (instead of the above).
268
+ ```
269
+ %run -m qai_hub_models.models.yolov8_det.demo -- --eval-mode on-device
270
+ ```
271
+
272
+
273
+ ## Deploying compiled model to Android
274
+
275
+
276
+ The models can be deployed using multiple runtimes:
277
+ - TensorFlow Lite (`.tflite` export): [This
278
+ tutorial](https://www.tensorflow.org/lite/android/quickstart) provides a
279
+ guide to deploy the .tflite model in an Android application.
280
+
281
+
282
+ - QNN (`.so` export ): This [sample
283
+ app](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/sample_app.html)
284
+ provides instructions on how to use the `.so` shared library in an Android application.
285
+
286
+
287
+ ## View on Qualcomm® AI Hub
288
+ Get more details on YOLOv8-Detection's performance across various devices [here](https://aihub.qualcomm.com/models/yolov8_det).
289
+ Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/)
290
 
291
 
292
  ## License
 
303
 
304
 
305
  ## Community
306
+ * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
307
  * For questions or feedback please [reach out to us](mailto:[email protected]).
308
 
309
+