Hugging Face's logo

qualcomm
/
DeepLabV3-ResNet50

Image Segmentation
PyTorch
LiteRT
android
Model card Files
xet
Community
bhushans commited on
Commit
30d9acb
·
verified ·
1 Parent(s): df37d26

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +29 -12
README.md CHANGED
@@ -35,15 +35,16 @@ More details on model performance across various devices, can be found
35
 
36
  | Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
37
  |---|---|---|---|---|---|---|---|---|
38
- | DeepLabV3-ResNet50 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 294.937 ms | 0 - 338 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
39
- | DeepLabV3-ResNet50 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 206.139 ms | 21 - 47 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
40
- | DeepLabV3-ResNet50 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 222.046 ms | 22 - 42 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
41
- | DeepLabV3-ResNet50 | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 292.298 ms | 0 - 268 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
42
- | DeepLabV3-ResNet50 | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 296.485 ms | 2 - 134 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
43
- | DeepLabV3-ResNet50 | SA8775 (Proxy) | SA8775P Proxy | TFLITE | 292.364 ms | 0 - 128 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
44
- | DeepLabV3-ResNet50 | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 295.452 ms | 0 - 164 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
45
- | DeepLabV3-ResNet50 | SA8295P ADP | SA8295P | TFLITE | 279.707 ms | 6 - 26 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
46
- | DeepLabV3-ResNet50 | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 467.317 ms | 23 - 51 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
 
47
 
48
 
49
 
@@ -107,8 +108,8 @@ Profiling Results
107
  DeepLabV3-ResNet50
108
  Device : Samsung Galaxy S23 (13)
109
  Runtime : TFLITE
110
- Estimated inference time (ms) : 294.9
111
- Estimated peak memory usage (MB): [0, 338]
112
  Total # Ops : 100
113
  Compute Unit(s) : GPU (98 ops) CPU (2 ops)
114
  ```
@@ -129,13 +130,29 @@ in memory using the `jit.trace` and then call the `submit_compile_job` API.
129
  import torch
130
 
131
  import qai_hub as hub
132
- from qai_hub_models.models.deeplabv3_resnet50 import
133
 
134
  # Load the model
 
135
 
136
  # Device
137
  device = hub.Device("Samsung Galaxy S23")
138
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
139
 
140
  ```
141
 
 
35
 
36
  | Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
37
  |---|---|---|---|---|---|---|---|---|
38
+ | DeepLabV3-ResNet50 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 290.866 ms | 0 - 164 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
39
+ | DeepLabV3-ResNet50 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 206.66 ms | 21 - 46 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
40
+ | DeepLabV3-ResNet50 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 216.408 ms | 12 - 28 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
41
+ | DeepLabV3-ResNet50 | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 291.878 ms | 0 - 142 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
42
+ | DeepLabV3-ResNet50 | SA7255P ADP | SA7255P | TFLITE | 2151.421 ms | 21 - 42 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
43
+ | DeepLabV3-ResNet50 | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 291.325 ms | 6 - 175 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
44
+ | DeepLabV3-ResNet50 | SA8295P ADP | SA8295P | TFLITE | 281.323 ms | 6 - 26 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
45
+ | DeepLabV3-ResNet50 | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 291.54 ms | 0 - 161 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
46
+ | DeepLabV3-ResNet50 | SA8775P ADP | SA8775P | TFLITE | 592.859 ms | 22 - 44 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
47
+ | DeepLabV3-ResNet50 | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 408.225 ms | 23 - 52 MB | FP16 | GPU | [DeepLabV3-ResNet50.tflite](https://huggingface.co/qualcomm/DeepLabV3-ResNet50/blob/main/DeepLabV3-ResNet50.tflite) |
48
 
49
 
50
 
 
108
  DeepLabV3-ResNet50
109
  Device : Samsung Galaxy S23 (13)
110
  Runtime : TFLITE
111
+ Estimated inference time (ms) : 290.9
112
+ Estimated peak memory usage (MB): [0, 164]
113
  Total # Ops : 100
114
  Compute Unit(s) : GPU (98 ops) CPU (2 ops)
115
  ```
 
130
  import torch
131
 
132
  import qai_hub as hub
133
+ from qai_hub_models.models.deeplabv3_resnet50 import Model
134
 
135
  # Load the model
136
+ torch_model = Model.from_pretrained()
137
 
138
  # Device
139
  device = hub.Device("Samsung Galaxy S23")
140
 
141
+ # Trace model
142
+ input_shape = torch_model.get_input_spec()
143
+ sample_inputs = torch_model.sample_inputs()
144
+
145
+ pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])
146
+
147
+ # Compile model on a specific device
148
+ compile_job = hub.submit_compile_job(
149
+ model=pt_model,
150
+ device=device,
151
+ input_specs=torch_model.get_input_spec(),
152
+ )
153
+
154
+ # Get target model to run on-device
155
+ target_model = compile_job.get_target_model()
156
 
157
  ```
158