Hugging Face's logo

qualcomm
/
Facial-Attribute-Detection-Quantized

Object Detection
PyTorch
LiteRT
ONNX
real_time
quantized
android
Model card Files Community
qaihm-bot commited on
Commit
0e87e6a
·
verified ·
1 Parent(s): 95bef7e

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +111 -22
README.md CHANGED
@@ -37,24 +37,38 @@ More details on model performance across various devices, can be found
37
 
38
  | Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
39
  |---|---|---|---|---|---|---|---|---|
40
- | Facial-Attribute-Detection-Quantized | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 0.434 ms | 0 - 51 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
41
- | Facial-Attribute-Detection-Quantized | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 0.942 ms | 0 - 41 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.onnx](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.onnx) |
42
- | Facial-Attribute-Detection-Quantized | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 0.33 ms | 0 - 44 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
43
- | Facial-Attribute-Detection-Quantized | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 0.716 ms | 0 - 59 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.onnx](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.onnx) |
44
- | Facial-Attribute-Detection-Quantized | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 0.356 ms | 0 - 42 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
45
- | Facial-Attribute-Detection-Quantized | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 0.598 ms | 0 - 44 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.onnx](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.onnx) |
46
- | Facial-Attribute-Detection-Quantized | SA7255P ADP | SA7255P | TFLITE | 4.839 ms | 0 - 30 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
47
- | Facial-Attribute-Detection-Quantized | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 0.434 ms | 0 - 51 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
48
- | Facial-Attribute-Detection-Quantized | SA8295P ADP | SA8295P | TFLITE | 0.929 ms | 0 - 36 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
49
- | Facial-Attribute-Detection-Quantized | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 0.434 ms | 0 - 52 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
50
- | Facial-Attribute-Detection-Quantized | SA8775P ADP | SA8775P | TFLITE | 0.79 ms | 0 - 31 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
51
- | Facial-Attribute-Detection-Quantized | RB3 Gen 2 (Proxy) | QCS6490 Proxy | TFLITE | 1.417 ms | 0 - 38 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
52
- | Facial-Attribute-Detection-Quantized | RB5 (Proxy) | QCS8250 Proxy | TFLITE | 81.993 ms | 2 - 5 MB | FP32 | CPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
53
- | Facial-Attribute-Detection-Quantized | QCS8275 (Proxy) | QCS8275 Proxy | TFLITE | 4.839 ms | 0 - 30 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
54
- | Facial-Attribute-Detection-Quantized | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 0.439 ms | 0 - 51 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
55
- | Facial-Attribute-Detection-Quantized | QCS9075 (Proxy) | QCS9075 Proxy | TFLITE | 0.79 ms | 0 - 31 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
56
- | Facial-Attribute-Detection-Quantized | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 0.564 ms | 0 - 42 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
57
- | Facial-Attribute-Detection-Quantized | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 0.927 ms | 13 - 13 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.onnx](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.onnx) |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
58
 
59
 
60
 
@@ -119,12 +133,88 @@ Facial-Attribute-Detection-Quantized
119
  Device : Samsung Galaxy S23 (13)
120
  Runtime : TFLITE
121
  Estimated inference time (ms) : 0.4
122
- Estimated peak memory usage (MB): [0, 51]
123
- Total # Ops : 168
124
- Compute Unit(s) : NPU (168 ops)
125
  ```
126
 
127
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
128
 
129
 
130
  ## Run demo on a cloud-hosted device
@@ -169,7 +259,6 @@ Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/)
169
 
170
 
171
  ## References
172
- * [None](None)
173
  * [Source Model Implementation](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/face_attrib_net/model.py)
174
 
175
 
 
37
 
38
  | Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
39
  |---|---|---|---|---|---|---|---|---|
40
+ | Facial-Attribute-Detection-Quantized | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 0.422 ms | 0 - 50 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
41
+ | Facial-Attribute-Detection-Quantized | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 0.418 ms | 0 - 3 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.so](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.so) |
42
+ | Facial-Attribute-Detection-Quantized | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 0.64 ms | 0 - 44 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.onnx](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.onnx) |
43
+ | Facial-Attribute-Detection-Quantized | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 0.323 ms | 0 - 45 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
44
+ | Facial-Attribute-Detection-Quantized | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 0.3 ms | 0 - 21 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.so](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.so) |
45
+ | Facial-Attribute-Detection-Quantized | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 0.436 ms | 0 - 54 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.onnx](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.onnx) |
46
+ | Facial-Attribute-Detection-Quantized | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 0.324 ms | 0 - 35 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
47
+ | Facial-Attribute-Detection-Quantized | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 0.327 ms | 0 - 32 MB | INT8 | NPU | Use Export Script |
48
+ | Facial-Attribute-Detection-Quantized | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 0.465 ms | 0 - 43 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.onnx](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.onnx) |
49
+ | Facial-Attribute-Detection-Quantized | SA7255P ADP | SA7255P | TFLITE | 4.933 ms | 0 - 31 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
50
+ | Facial-Attribute-Detection-Quantized | SA7255P ADP | SA7255P | QNN | 4.857 ms | 0 - 10 MB | INT8 | NPU | Use Export Script |
51
+ | Facial-Attribute-Detection-Quantized | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 0.427 ms | 0 - 50 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
52
+ | Facial-Attribute-Detection-Quantized | SA8255 (Proxy) | SA8255P Proxy | QNN | 0.418 ms | 0 - 3 MB | INT8 | NPU | Use Export Script |
53
+ | Facial-Attribute-Detection-Quantized | SA8295P ADP | SA8295P | TFLITE | 0.929 ms | 0 - 34 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
54
+ | Facial-Attribute-Detection-Quantized | SA8295P ADP | SA8295P | QNN | 0.845 ms | 0 - 18 MB | INT8 | NPU | Use Export Script |
55
+ | Facial-Attribute-Detection-Quantized | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 0.419 ms | 0 - 51 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
56
+ | Facial-Attribute-Detection-Quantized | SA8650 (Proxy) | SA8650P Proxy | QNN | 0.421 ms | 0 - 3 MB | INT8 | NPU | Use Export Script |
57
+ | Facial-Attribute-Detection-Quantized | SA8775P ADP | SA8775P | TFLITE | 0.767 ms | 0 - 31 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
58
+ | Facial-Attribute-Detection-Quantized | SA8775P ADP | SA8775P | QNN | 0.701 ms | 0 - 10 MB | INT8 | NPU | Use Export Script |
59
+ | Facial-Attribute-Detection-Quantized | RB3 Gen 2 (Proxy) | QCS6490 Proxy | TFLITE | 1.46 ms | 0 - 41 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
60
+ | Facial-Attribute-Detection-Quantized | RB3 Gen 2 (Proxy) | QCS6490 Proxy | QNN | 1.564 ms | 0 - 14 MB | INT8 | NPU | Use Export Script |
61
+ | Facial-Attribute-Detection-Quantized | RB5 (Proxy) | QCS8250 Proxy | TFLITE | 69.358 ms | 2 - 5 MB | FP32 | CPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
62
+ | Facial-Attribute-Detection-Quantized | QCS8275 (Proxy) | QCS8275 Proxy | TFLITE | 4.933 ms | 0 - 31 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
63
+ | Facial-Attribute-Detection-Quantized | QCS8275 (Proxy) | QCS8275 Proxy | QNN | 4.857 ms | 0 - 10 MB | INT8 | NPU | Use Export Script |
64
+ | Facial-Attribute-Detection-Quantized | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 0.422 ms | 0 - 49 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
65
+ | Facial-Attribute-Detection-Quantized | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 0.417 ms | 0 - 2 MB | INT8 | NPU | Use Export Script |
66
+ | Facial-Attribute-Detection-Quantized | QCS9075 (Proxy) | QCS9075 Proxy | TFLITE | 0.767 ms | 0 - 31 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
67
+ | Facial-Attribute-Detection-Quantized | QCS9075 (Proxy) | QCS9075 Proxy | QNN | 0.701 ms | 0 - 10 MB | INT8 | NPU | Use Export Script |
68
+ | Facial-Attribute-Detection-Quantized | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 0.669 ms | 0 - 45 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.tflite](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.tflite) |
69
+ | Facial-Attribute-Detection-Quantized | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 0.65 ms | 0 - 44 MB | INT8 | NPU | Use Export Script |
70
+ | Facial-Attribute-Detection-Quantized | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 0.513 ms | 0 - 0 MB | INT8 | NPU | Use Export Script |
71
+ | Facial-Attribute-Detection-Quantized | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 0.609 ms | 13 - 13 MB | INT8 | NPU | [Facial-Attribute-Detection-Quantized.onnx](https://huggingface.co/qualcomm/Facial-Attribute-Detection-Quantized/blob/main/Facial-Attribute-Detection-Quantized.onnx) |
72
 
73
 
74
 
 
133
  Device : Samsung Galaxy S23 (13)
134
  Runtime : TFLITE
135
  Estimated inference time (ms) : 0.4
136
+ Estimated peak memory usage (MB): [0, 50]
137
+ Total # Ops : 171
138
+ Compute Unit(s) : NPU (171 ops)
139
  ```
140
 
141
 
142
+ ## How does this work?
143
+
144
+ This [export script](https://aihub.qualcomm.com/models/face_attrib_net_quantized/qai_hub_models/models/Facial-Attribute-Detection-Quantized/export.py)
145
+ leverages [Qualcomm® AI Hub](https://aihub.qualcomm.com/) to optimize, validate, and deploy this model
146
+ on-device. Lets go through each step below in detail:
147
+
148
+ Step 1: **Compile model for on-device deployment**
149
+
150
+ To compile a PyTorch model for on-device deployment, we first trace the model
151
+ in memory using the `jit.trace` and then call the `submit_compile_job` API.
152
+
153
+ ```python
154
+ import torch
155
+
156
+ import qai_hub as hub
157
+ from qai_hub_models.models.face_attrib_net_quantized import Model
158
+
159
+ # Load the model
160
+ torch_model = Model.from_pretrained()
161
+
162
+ # Device
163
+ device = hub.Device("Samsung Galaxy S24")
164
+
165
+ # Trace model
166
+ input_shape = torch_model.get_input_spec()
167
+ sample_inputs = torch_model.sample_inputs()
168
+
169
+ pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])
170
+
171
+ # Compile model on a specific device
172
+ compile_job = hub.submit_compile_job(
173
+ model=pt_model,
174
+ device=device,
175
+ input_specs=torch_model.get_input_spec(),
176
+ )
177
+
178
+ # Get target model to run on-device
179
+ target_model = compile_job.get_target_model()
180
+
181
+ ```
182
+
183
+
184
+ Step 2: **Performance profiling on cloud-hosted device**
185
+
186
+ After compiling models from step 1. Models can be profiled model on-device using the
187
+ `target_model`. Note that this scripts runs the model on a device automatically
188
+ provisioned in the cloud. Once the job is submitted, you can navigate to a
189
+ provided job URL to view a variety of on-device performance metrics.
190
+ ```python
191
+ profile_job = hub.submit_profile_job(
192
+ model=target_model,
193
+ device=device,
194
+ )
195
+
196
+ ```
197
+
198
+ Step 3: **Verify on-device accuracy**
199
+
200
+ To verify the accuracy of the model on-device, you can run on-device inference
201
+ on sample input data on the same cloud hosted device.
202
+ ```python
203
+ input_data = torch_model.sample_inputs()
204
+ inference_job = hub.submit_inference_job(
205
+ model=target_model,
206
+ device=device,
207
+ inputs=input_data,
208
+ )
209
+ on_device_output = inference_job.download_output_data()
210
+
211
+ ```
212
+ With the output of the model, you can compute like PSNR, relative errors or
213
+ spot check the output with expected output.
214
+
215
+ **Note**: This on-device profiling and inference requires access to Qualcomm®
216
+ AI Hub. [Sign up for access](https://myaccount.qualcomm.com/signup).
217
+
218
 
219
 
220
  ## Run demo on a cloud-hosted device
 
259
 
260
 
261
  ## References
 
262
  * [Source Model Implementation](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/face_attrib_net/model.py)
263
 
264