Hugging Face's logo

qualcomm
/
DETR-ResNet50-DC5

Object Detection
PyTorch
TF Lite
ONNX
android
Model card Files Community
qaihm-bot commited on
Commit
98b4316
·
verified ·
1 Parent(s): 4b86b9e

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +33 -16
README.md CHANGED
@@ -34,19 +34,20 @@ More details on model performance across various devices, can be found
34
 
35
  | Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
36
  |---|---|---|---|---|---|---|---|---|
37
- | DETR-ResNet50-DC5 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 74.27 ms | 0 - 2 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
38
- | DETR-ResNet50-DC5 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 89.233 ms | 0 - 95 MB | FP16 | NPU | [DETR-ResNet50-DC5.onnx](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.onnx) |
39
- | DETR-ResNet50-DC5 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 59.252 ms | 0 - 493 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
40
- | DETR-ResNet50-DC5 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 76.827 ms | 3 - 506 MB | FP16 | NPU | [DETR-ResNet50-DC5.onnx](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.onnx) |
41
- | DETR-ResNet50-DC5 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 55.125 ms | 0 - 253 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
42
- | DETR-ResNet50-DC5 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 62.899 ms | 0 - 292 MB | FP16 | NPU | [DETR-ResNet50-DC5.onnx](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.onnx) |
43
- | DETR-ResNet50-DC5 | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 74.483 ms | 0 - 3 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
44
- | DETR-ResNet50-DC5 | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 83.592 ms | 0 - 2 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
45
- | DETR-ResNet50-DC5 | SA8775 (Proxy) | SA8775P Proxy | TFLITE | 82.707 ms | 0 - 2 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
46
- | DETR-ResNet50-DC5 | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 75.931 ms | 0 - 2 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
47
- | DETR-ResNet50-DC5 | SA8295P ADP | SA8295P | TFLITE | 92.792 ms | 0 - 224 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
48
- | DETR-ResNet50-DC5 | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 91.722 ms | 0 - 455 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
49
- | DETR-ResNet50-DC5 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 65.268 ms | 84 - 84 MB | FP16 | NPU | [DETR-ResNet50-DC5.onnx](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.onnx) |
 
50
 
51
 
52
 
@@ -111,8 +112,8 @@ Profiling Results
111
  DETR-ResNet50-DC5
112
  Device : Samsung Galaxy S23 (13)
113
  Runtime : TFLITE
114
- Estimated inference time (ms) : 74.3
115
- Estimated peak memory usage (MB): [0, 2]
116
  Total # Ops : 789
117
  Compute Unit(s) : NPU (789 ops)
118
  ```
@@ -133,13 +134,29 @@ in memory using the `jit.trace` and then call the `submit_compile_job` API.
133
  import torch
134
 
135
  import qai_hub as hub
136
- from qai_hub_models.models.detr_resnet50_dc5 import
137
 
138
  # Load the model
 
139
 
140
  # Device
141
  device = hub.Device("Samsung Galaxy S23")
142
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
143
 
144
  ```
145
 
 
34
 
35
  | Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
36
  |---|---|---|---|---|---|---|---|---|
37
+ | DETR-ResNet50-DC5 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 84.968 ms | 0 - 46 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
38
+ | DETR-ResNet50-DC5 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 70.957 ms | 0 - 96 MB | FP16 | NPU | [DETR-ResNet50-DC5.onnx](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.onnx) |
39
+ | DETR-ResNet50-DC5 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 60.654 ms | 0 - 148 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
40
+ | DETR-ResNet50-DC5 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 53.404 ms | 3 - 476 MB | FP16 | NPU | [DETR-ResNet50-DC5.onnx](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.onnx) |
41
+ | DETR-ResNet50-DC5 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 49.391 ms | 0 - 155 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
42
+ | DETR-ResNet50-DC5 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 44.861 ms | 0 - 269 MB | FP16 | NPU | [DETR-ResNet50-DC5.onnx](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.onnx) |
43
+ | DETR-ResNet50-DC5 | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 85.054 ms | 0 - 51 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
44
+ | DETR-ResNet50-DC5 | SA7255P ADP | SA7255P | TFLITE | 631.553 ms | 0 - 155 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
45
+ | DETR-ResNet50-DC5 | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 84.991 ms | 0 - 45 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
46
+ | DETR-ResNet50-DC5 | SA8295P ADP | SA8295P | TFLITE | 94.835 ms | 0 - 137 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
47
+ | DETR-ResNet50-DC5 | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 86.776 ms | 0 - 54 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
48
+ | DETR-ResNet50-DC5 | SA8775P ADP | SA8775P | TFLITE | 95.923 ms | 0 - 154 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
49
+ | DETR-ResNet50-DC5 | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 98.287 ms | 0 - 135 MB | FP16 | NPU | [DETR-ResNet50-DC5.tflite](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.tflite) |
50
+ | DETR-ResNet50-DC5 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 65.507 ms | 83 - 83 MB | FP16 | NPU | [DETR-ResNet50-DC5.onnx](https://huggingface.co/qualcomm/DETR-ResNet50-DC5/blob/main/DETR-ResNet50-DC5.onnx) |
51
 
52
 
53
 
 
112
  DETR-ResNet50-DC5
113
  Device : Samsung Galaxy S23 (13)
114
  Runtime : TFLITE
115
+ Estimated inference time (ms) : 85.0
116
+ Estimated peak memory usage (MB): [0, 46]
117
  Total # Ops : 789
118
  Compute Unit(s) : NPU (789 ops)
119
  ```
 
134
  import torch
135
 
136
  import qai_hub as hub
137
+ from qai_hub_models.models.detr_resnet50_dc5 import Model
138
 
139
  # Load the model
140
+ torch_model = Model.from_pretrained()
141
 
142
  # Device
143
  device = hub.Device("Samsung Galaxy S23")
144
 
145
+ # Trace model
146
+ input_shape = torch_model.get_input_spec()
147
+ sample_inputs = torch_model.sample_inputs()
148
+
149
+ pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])
150
+
151
+ # Compile model on a specific device
152
+ compile_job = hub.submit_compile_job(
153
+ model=pt_model,
154
+ device=device,
155
+ input_specs=torch_model.get_input_spec(),
156
+ )
157
+
158
+ # Get target model to run on-device
159
+ target_model = compile_job.get_target_model()
160
 
161
  ```
162