Hugging Face's logo

qualcomm
/
SwinV2-Base

Image Classification
PyTorch
backbone
bu_auto
android
Model card Files
xet
 Community
qaihm-bot commited on
Commit
0f9ebc7
·
verified ·
1 Parent(s): 5228d55

v0.60.0

Browse files

See https://github.com/qualcomm/ai-hub-models/releases/v0.60.0 for changelog.

Files changed (2) hide show
  1. README.md +55 -53
  2. release_assets.json +4 -4
README.md CHANGED
@@ -16,7 +16,7 @@ pipeline_tag: image-classification
16
  SwinV2Base is a machine learning model that can classify images from the Imagenet dataset. It can also be used as a backbone in building more complex models for specific use cases.
17
 
18
  This is based on the implementation of SwinV2-Base found [here](https://github.com/pytorch/vision/blob/main/torchvision/models/swin_transformer.py).
19
- This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/swinv2_base) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).
20
 
21
  Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device.
22
 
@@ -29,83 +29,85 @@ Below are pre-exported model assets ready for deployment.
29
 
30
  | Runtime | Precision | Chipset | SDK Versions | Download |
31
  |---|---|---|---|---|
32
- | QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swinv2_base/releases/v0.59.0/swinv2_base-qnn_dlc-float.zip)
33
- | QNN_DLC | w8a16 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swinv2_base/releases/v0.59.0/swinv2_base-qnn_dlc-w8a16.zip)
34
- | TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swinv2_base/releases/v0.59.0/swinv2_base-tflite-float.zip)
35
 
36
  For more device-specific assets and performance metrics, visit **[SwinV2-Base on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/swinv2_base)**.
37
 
38
 
39
  ### Option 2: Export with Custom Configurations
40
 
41
- Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/swinv2_base) Python library to compile and export the model with your own:
42
  - Custom weights (e.g., fine-tuned checkpoints)
43
  - Custom input shapes
44
  - Target device and runtime configurations
45
 
46
  This option is ideal if you need to customize the model beyond the default configuration provided here.
47
 
48
- See our repository for [SwinV2-Base on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/swinv2_base) for usage instructions.
49
 
50
  ## Model Details
51
 
52
  **Model Type:** Model_use_case.image_classification
53
 
54
  **Model Stats:**
55
- - Model checkpoint: Imagenet
56
  - Input resolution: 256x256
57
- - Number of parameters: 88.8M
58
  - Model size (float): 339 MB
59
  - Model size (w8a16): 90.2 MB
 
60
 
61
  ## Performance Summary
62
  | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
63
  |---|---|---|---|---|---|---
64
- | SwinV2-Base | QNN_DLC | float | Snapdragon® X2 Elite | 12.451 ms | 1 - 1 MB | NPU
65
- | SwinV2-Base | QNN_DLC | float | Snapdragon® X Elite | 28.955 ms | 1 - 1 MB | NPU
66
- | SwinV2-Base | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 19.714 ms | 1 - 545 MB | NPU
67
- | SwinV2-Base | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 41.129 ms | 0 - 534 MB | NPU
68
- | SwinV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8275 | 74.071 ms | 1 - 395 MB | NPU
69
- | SwinV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 27.948 ms | 1 - 449 MB | NPU
70
- | SwinV2-Base | QNN_DLC | float | Qualcomm® SA8775P | 31.464 ms | 1 - 388 MB | NPU
71
- | SwinV2-Base | QNN_DLC | float | Qualcomm® SA8650P | 31.464 ms | 1 - 388 MB | NPU
72
- | SwinV2-Base | QNN_DLC | float | Qualcomm® SA8255P | 31.464 ms | 1 - 388 MB | NPU
73
- | SwinV2-Base | QNN_DLC | float | Qualcomm® QCS8450 | 41.129 ms | 0 - 534 MB | NPU
74
- | SwinV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 39.558 ms | 3 - 5 MB | NPU
75
- | SwinV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 28.955 ms | 1 - 1 MB | NPU
76
- | SwinV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 14.613 ms | 1 - 392 MB | NPU
77
- | SwinV2-Base | QNN_DLC | float | Qualcomm® SA7255P | 74.071 ms | 1 - 395 MB | NPU
78
- | SwinV2-Base | QNN_DLC | float | Qualcomm® SA8295P | 37.795 ms | 1 - 378 MB | NPU
79
- | SwinV2-Base | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 14.613 ms | 1 - 392 MB | NPU
80
- | SwinV2-Base | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 11.583 ms | 0 - 424 MB | NPU
81
- | SwinV2-Base | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 12.215 ms | 0 - 0 MB | NPU
82
- | SwinV2-Base | QNN_DLC | w8a16 | Snapdragon® X Elite | 30.702 ms | 0 - 0 MB | NPU
83
- | SwinV2-Base | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 19.654 ms | 0 - 2036 MB | NPU
84
- | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8275 | 52.667 ms | 0 - 486 MB | NPU
85
- | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 29.334 ms | 0 - 3 MB | NPU
86
- | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® SA8775P | 29.997 ms | 0 - 869 MB | NPU
87
- | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® SA8650P | 29.997 ms | 0 - 869 MB | NPU
88
- | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® SA8255P | 29.997 ms | 0 - 869 MB | NPU
89
- | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 29.947 ms | 0 - 2 MB | NPU
90
- | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 30.702 ms | 0 - 0 MB | NPU
91
- | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 14.736 ms | 0 - 903 MB | NPU
92
- | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® SA7255P | 52.667 ms | 0 - 486 MB | NPU
93
- | SwinV2-Base | QNN_DLC | w8a16 | Snapdragon® 8 Elite Mobile | 14.736 ms | 0 - 903 MB | NPU
94
- | SwinV2-Base | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 11.415 ms | 0 - 953 MB | NPU
95
- | SwinV2-Base | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 19.963 ms | 0 - 2174 MB | NPU
96
- | SwinV2-Base | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 42.639 ms | 0 - 676 MB | NPU
97
- | SwinV2-Base | TFLITE | float | Qualcomm® Dragonwing™ QCS8275 | 71.105 ms | 0 - 885 MB | NPU
98
- | SwinV2-Base | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 29.556 ms | 0 - 4 MB | NPU
99
- | SwinV2-Base | TFLITE | float | Qualcomm® SA8775P | 32.555 ms | 0 - 879 MB | NPU
100
- | SwinV2-Base | TFLITE | float | Qualcomm® SA8650P | 32.555 ms | 0 - 879 MB | NPU
101
- | SwinV2-Base | TFLITE | float | Qualcomm® SA8255P | 32.555 ms | 0 - 879 MB | NPU
102
- | SwinV2-Base | TFLITE | float | Qualcomm® QCS8450 | 42.639 ms | 0 - 676 MB | NPU
103
- | SwinV2-Base | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 32.565 ms | 0 - 181 MB | NPU
104
- | SwinV2-Base | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 14.878 ms | 0 - 896 MB | NPU
105
- | SwinV2-Base | TFLITE | float | Qualcomm® SA7255P | 71.105 ms | 0 - 885 MB | NPU
106
- | SwinV2-Base | TFLITE | float | Qualcomm® SA8295P | 40.584 ms | 0 - 875 MB | NPU
107
- | SwinV2-Base | TFLITE | float | Snapdragon® 8 Elite Mobile | 14.878 ms | 0 - 896 MB | NPU
108
- | SwinV2-Base | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 11.386 ms | 0 - 939 MB | NPU
 
 
109
 
110
  ## License
111
  * The license for the original implementation of SwinV2-Base can be found
 
16
  SwinV2Base is a machine learning model that can classify images from the Imagenet dataset. It can also be used as a backbone in building more complex models for specific use cases.
17
 
18
  This is based on the implementation of SwinV2-Base found [here](https://github.com/pytorch/vision/blob/main/torchvision/models/swin_transformer.py).
19
+ This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.60.0/src/qai_hub_models/models/swinv2_base) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).
20
 
21
  Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device.
22
 
 
29
 
30
  | Runtime | Precision | Chipset | SDK Versions | Download |
31
  |---|---|---|---|---|
32
+ | QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swinv2_base/releases/v0.60.0/swinv2_base-qnn_dlc-float.zip)
33
+ | QNN_DLC | w8a16 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swinv2_base/releases/v0.60.0/swinv2_base-qnn_dlc-w8a16.zip)
34
+ | TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swinv2_base/releases/v0.60.0/swinv2_base-tflite-float.zip)
35
 
36
  For more device-specific assets and performance metrics, visit **[SwinV2-Base on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/swinv2_base)**.
37
 
38
 
39
  ### Option 2: Export with Custom Configurations
40
 
41
+ Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.60.0/src/qai_hub_models/models/swinv2_base) Python library to compile and export the model with your own:
42
  - Custom weights (e.g., fine-tuned checkpoints)
43
  - Custom input shapes
44
  - Target device and runtime configurations
45
 
46
  This option is ideal if you need to customize the model beyond the default configuration provided here.
47
 
48
+ See our repository for [SwinV2-Base on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.60.0/src/qai_hub_models/models/swinv2_base) for usage instructions.
49
 
50
  ## Model Details
51
 
52
  **Model Type:** Model_use_case.image_classification
53
 
54
  **Model Stats:**
 
55
  - Input resolution: 256x256
56
+ - Model checkpoint: Imagenet
57
  - Model size (float): 339 MB
58
  - Model size (w8a16): 90.2 MB
59
+ - Number of parameters: 88.8M
60
 
61
  ## Performance Summary
62
  | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
63
  |---|---|---|---|---|---|---
64
+ | SwinV2-Base | QNN_DLC | float | Snapdragon® X2 Elite | 12.473 ms | 1 - 1 MB | NPU
65
+ | SwinV2-Base | QNN_DLC | float | Snapdragon® X Elite | 28.971 ms | 1 - 1 MB | NPU
66
+ | SwinV2-Base | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 19.578 ms | 0 - 545 MB | NPU
67
+ | SwinV2-Base | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 41.304 ms | 0 - 531 MB | NPU
68
+ | SwinV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 30.44 ms | 1 - 4 MB | NPU
69
+ | SwinV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 74.085 ms | 1 - 395 MB | NPU
70
+ | SwinV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 27.695 ms | 1 - 3 MB | NPU
71
+ | SwinV2-Base | QNN_DLC | float | Qualcomm® SA8775P | 31.457 ms | 1 - 388 MB | NPU
72
+ | SwinV2-Base | QNN_DLC | float | Qualcomm® SA8650P | 31.457 ms | 1 - 388 MB | NPU
73
+ | SwinV2-Base | QNN_DLC | float | Qualcomm® SA8255P | 31.457 ms | 1 - 388 MB | NPU
74
+ | SwinV2-Base | QNN_DLC | float | Qualcomm® QCS8450 | 41.304 ms | 0 - 531 MB | NPU
75
+ | SwinV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 31.518 ms | 3 - 5 MB | NPU
76
+ | SwinV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 28.971 ms | 1 - 1 MB | NPU
77
+ | SwinV2-Base | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 14.734 ms | 1 - 391 MB | NPU
78
+ | SwinV2-Base | QNN_DLC | float | Qualcomm® SA7255P | 74.085 ms | 1 - 395 MB | NPU
79
+ | SwinV2-Base | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 14.734 ms | 1 - 391 MB | NPU
80
+ | SwinV2-Base | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 11.594 ms | 0 - 425 MB | NPU
81
+ | SwinV2-Base | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 12.159 ms | 0 - 0 MB | NPU
82
+ | SwinV2-Base | QNN_DLC | w8a16 | Snapdragon® X Elite | 30.727 ms | 0 - 0 MB | NPU
83
+ | SwinV2-Base | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 19.643 ms | 0 - 2035 MB | NPU
84
+ | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-8275 | 26.218 ms | 0 - 3 MB | NPU
85
+ | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-8275 | 52.548 ms | 0 - 910 MB | NPU
86
+ | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 29.404 ms | 0 - 2 MB | NPU
87
+ | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® SA8775P | 30.008 ms | 0 - 869 MB | NPU
88
+ | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® SA8650P | 30.008 ms | 0 - 869 MB | NPU
89
+ | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® SA8255P | 30.008 ms | 0 - 869 MB | NPU
90
+ | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 30.253 ms | 0 - 2 MB | NPU
91
+ | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 30.727 ms | 0 - 0 MB | NPU
92
+ | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 14.804 ms | 0 - 904 MB | NPU
93
+ | SwinV2-Base | QNN_DLC | w8a16 | Qualcomm® SA7255P | 52.548 ms | 0 - 910 MB | NPU
94
+ | SwinV2-Base | QNN_DLC | w8a16 | Snapdragon® 8 Elite Mobile | 14.804 ms | 0 - 904 MB | NPU
95
+ | SwinV2-Base | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 11.418 ms | 0 - 953 MB | NPU
96
+ | SwinV2-Base | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 20.065 ms | 0 - 2179 MB | NPU
97
+ | SwinV2-Base | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 42.481 ms | 0 - 678 MB | NPU
98
+ | SwinV2-Base | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 30.064 ms | 0 - 181 MB | NPU
99
+ | SwinV2-Base | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 71.097 ms | 0 - 885 MB | NPU
100
+ | SwinV2-Base | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 29.508 ms | 0 - 4 MB | NPU
101
+ | SwinV2-Base | TFLITE | float | Qualcomm® SA8775P | 32.542 ms | 0 - 879 MB | NPU
102
+ | SwinV2-Base | TFLITE | float | Qualcomm® SA8650P | 32.542 ms | 0 - 879 MB | NPU
103
+ | SwinV2-Base | TFLITE | float | Qualcomm® SA8255P | 32.542 ms | 0 - 879 MB | NPU
104
+ | SwinV2-Base | TFLITE | float | Qualcomm® QCS8450 | 42.481 ms | 0 - 678 MB | NPU
105
+ | SwinV2-Base | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 32.411 ms | 0 - 181 MB | NPU
106
+ | SwinV2-Base | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 14.959 ms | 0 - 901 MB | NPU
107
+ | SwinV2-Base | TFLITE | float | Qualcomm® SA7255P | 71.097 ms | 0 - 885 MB | NPU
108
+ | SwinV2-Base | TFLITE | float | Qualcomm® SA8295P | 40.62 ms | 0 - 875 MB | NPU
109
+ | SwinV2-Base | TFLITE | float | Snapdragon® 8 Elite Mobile | 14.959 ms | 0 - 901 MB | NPU
110
+ | SwinV2-Base | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 11.388 ms | 0 - 939 MB | NPU
111
 
112
  ## License
113
  * The license for the original implementation of SwinV2-Base can be found
release_assets.json CHANGED
@@ -1,5 +1,5 @@
1
  {
2
- "version": "0.59.0",
3
  "precisions": {
4
  "float": {
5
  "universal_assets": {
@@ -7,13 +7,13 @@
7
  "tool_versions": {
8
  "qairt": "2.45.0.260326154327"
9
  },
10
- "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swinv2_base/releases/v0.59.0/swinv2_base-qnn_dlc-float.zip"
11
  },
12
  "tflite": {
13
  "tool_versions": {
14
  "qairt": "2.45.0.260326154327"
15
  },
16
- "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swinv2_base/releases/v0.59.0/swinv2_base-tflite-float.zip"
17
  }
18
  }
19
  },
@@ -23,7 +23,7 @@
23
  "tool_versions": {
24
  "qairt": "2.45.0.260326154327"
25
  },
26
- "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swinv2_base/releases/v0.59.0/swinv2_base-qnn_dlc-w8a16.zip"
27
  }
28
  }
29
  }
 
1
  {
2
+ "version": "0.60.0",
3
  "precisions": {
4
  "float": {
5
  "universal_assets": {
 
7
  "tool_versions": {
8
  "qairt": "2.45.0.260326154327"
9
  },
10
+ "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swinv2_base/releases/v0.60.0/swinv2_base-qnn_dlc-float.zip"
11
  },
12
  "tflite": {
13
  "tool_versions": {
14
  "qairt": "2.45.0.260326154327"
15
  },
16
+ "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swinv2_base/releases/v0.60.0/swinv2_base-tflite-float.zip"
17
  }
18
  }
19
  },
 
23
  "tool_versions": {
24
  "qairt": "2.45.0.260326154327"
25
  },
26
+ "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/swinv2_base/releases/v0.60.0/swinv2_base-qnn_dlc-w8a16.zip"
27
  }
28
  }
29
  }