Hugging Face's logo

qualcomm
/
Midas-V2

Depth Estimation
PyTorch
LiteRT
ONNX
android
Model card Files Community
qaihm-bot commited on
Commit
45120b5
·
verified ·
1 Parent(s): 0794157

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +40 -19
README.md CHANGED
@@ -14,7 +14,7 @@ tags:
14
 
15
  Midas is designed for estimating depth at each point in an image.
16
 
17
- This model is an implementation of Midas-V2 found [here](https://github.com/isl-org/MiDaS).
18
  This repository provides scripts to run Midas-V2 on Qualcomm® devices.
19
  More details on model performance across various devices, can be found
20
  [here](https://aihub.qualcomm.com/models/midas).
@@ -29,15 +29,32 @@ More details on model performance across various devices, can be found
29
  - Number of parameters: 16.6M
30
  - Model size: 63.2 MB
31
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
32
 
33
 
34
 
35
- | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
36
- | ---|---|---|---|---|---|---|---|
37
- | Samsung Galaxy S23 Ultra (Android 13) | Snapdragon® 8 Gen 2 | TFLite | 3.254 ms | 0 - 2 MB | FP16 | NPU | [Midas-V2.tflite](https://huggingface.co/qualcomm/Midas-V2/blob/main/Midas-V2.tflite)
38
- | Samsung Galaxy S23 Ultra (Android 13) | Snapdragon® 8 Gen 2 | QNN Model Library | 3.305 ms | 0 - 105 MB | FP16 | NPU | [Midas-V2.so](https://huggingface.co/qualcomm/Midas-V2/blob/main/Midas-V2.so)
39
-
40
-
41
 
42
  ## Installation
43
 
@@ -93,16 +110,16 @@ device. This script does the following:
93
  ```bash
94
  python -m qai_hub_models.models.midas.export
95
  ```
96
-
97
  ```
98
- Profile Job summary of Midas-V2
99
- --------------------------------------------------
100
- Device: Snapdragon X Elite CRD (11)
101
- Estimated Inference Time: 3.28 ms
102
- Estimated Peak Memory Range: 0.75-0.75 MB
103
- Compute Units: NPU (197) | Total (197)
104
-
105
-
 
106
  ```
107
 
108
 
@@ -201,15 +218,19 @@ provides instructions on how to use the `.so` shared library in an Android appl
201
  Get more details on Midas-V2's performance across various devices [here](https://aihub.qualcomm.com/models/midas).
202
  Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/)
203
 
 
204
  ## License
205
- - The license for the original implementation of Midas-V2 can be found
206
- [here](https://github.com/isl-org/MiDaS/blob/master/LICENSE).
207
- - The license for the compiled assets for on-device deployment can be found [here](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf)
 
208
 
209
  ## References
210
  * [Towards Robust Monocular Depth Estimation: Mixing Datasets for Zero-shot Cross-dataset Transfer](https://arxiv.org/abs/1907.01341v3)
211
  * [Source Model Implementation](https://github.com/isl-org/MiDaS)
212
 
 
 
213
  ## Community
214
  * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
215
  * For questions or feedback please [reach out to us](mailto:[email protected]).
 
14
 
15
  Midas is designed for estimating depth at each point in an image.
16
 
17
+ This model is an implementation of Midas-V2 found [here]({source_repo}).
18
  This repository provides scripts to run Midas-V2 on Qualcomm® devices.
19
  More details on model performance across various devices, can be found
20
  [here](https://aihub.qualcomm.com/models/midas).
 
29
  - Number of parameters: 16.6M
30
  - Model size: 63.2 MB
31
 
32
+ | Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
33
+ |---|---|---|---|---|---|---|---|---|
34
+ | Midas-V2 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 3.24 ms | 0 - 2 MB | FP16 | NPU | [Midas-V2.tflite](https://huggingface.co/qualcomm/Midas-V2/blob/main/Midas-V2.tflite) |
35
+ | Midas-V2 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 3.278 ms | 0 - 101 MB | FP16 | NPU | [Midas-V2.so](https://huggingface.co/qualcomm/Midas-V2/blob/main/Midas-V2.so) |
36
+ | Midas-V2 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 3.303 ms | 0 - 41 MB | FP16 | NPU | [Midas-V2.onnx](https://huggingface.co/qualcomm/Midas-V2/blob/main/Midas-V2.onnx) |
37
+ | Midas-V2 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 2.841 ms | 0 - 87 MB | FP16 | NPU | [Midas-V2.tflite](https://huggingface.co/qualcomm/Midas-V2/blob/main/Midas-V2.tflite) |
38
+ | Midas-V2 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 2.462 ms | 1 - 27 MB | FP16 | NPU | [Midas-V2.so](https://huggingface.co/qualcomm/Midas-V2/blob/main/Midas-V2.so) |
39
+ | Midas-V2 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 2.55 ms | 0 - 91 MB | FP16 | NPU | [Midas-V2.onnx](https://huggingface.co/qualcomm/Midas-V2/blob/main/Midas-V2.onnx) |
40
+ | Midas-V2 | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 3.213 ms | 0 - 5 MB | FP16 | NPU | [Midas-V2.tflite](https://huggingface.co/qualcomm/Midas-V2/blob/main/Midas-V2.tflite) |
41
+ | Midas-V2 | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 3.087 ms | 1 - 2 MB | FP16 | NPU | Use Export Script |
42
+ | Midas-V2 | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 3.222 ms | 0 - 2 MB | FP16 | NPU | [Midas-V2.tflite](https://huggingface.co/qualcomm/Midas-V2/blob/main/Midas-V2.tflite) |
43
+ | Midas-V2 | SA8255 (Proxy) | SA8255P Proxy | QNN | 3.045 ms | 1 - 2 MB | FP16 | NPU | Use Export Script |
44
+ | Midas-V2 | SA8775 (Proxy) | SA8775P Proxy | TFLITE | 3.228 ms | 0 - 2 MB | FP16 | NPU | [Midas-V2.tflite](https://huggingface.co/qualcomm/Midas-V2/blob/main/Midas-V2.tflite) |
45
+ | Midas-V2 | SA8775 (Proxy) | SA8775P Proxy | QNN | 3.049 ms | 1 - 2 MB | FP16 | NPU | Use Export Script |
46
+ | Midas-V2 | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 3.228 ms | 0 - 2 MB | FP16 | NPU | [Midas-V2.tflite](https://huggingface.co/qualcomm/Midas-V2/blob/main/Midas-V2.tflite) |
47
+ | Midas-V2 | SA8650 (Proxy) | SA8650P Proxy | QNN | 3.049 ms | 1 - 2 MB | FP16 | NPU | Use Export Script |
48
+ | Midas-V2 | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 4.752 ms | 0 - 91 MB | FP16 | NPU | [Midas-V2.tflite](https://huggingface.co/qualcomm/Midas-V2/blob/main/Midas-V2.tflite) |
49
+ | Midas-V2 | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 4.887 ms | 1 - 27 MB | FP16 | NPU | Use Export Script |
50
+ | Midas-V2 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 2.133 ms | 0 - 38 MB | FP16 | NPU | [Midas-V2.tflite](https://huggingface.co/qualcomm/Midas-V2/blob/main/Midas-V2.tflite) |
51
+ | Midas-V2 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 2.164 ms | 0 - 22 MB | FP16 | NPU | Use Export Script |
52
+ | Midas-V2 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 2.218 ms | 0 - 42 MB | FP16 | NPU | [Midas-V2.onnx](https://huggingface.co/qualcomm/Midas-V2/blob/main/Midas-V2.onnx) |
53
+ | Midas-V2 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 3.256 ms | 1 - 1 MB | FP16 | NPU | Use Export Script |
54
+ | Midas-V2 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 3.378 ms | 36 - 36 MB | FP16 | NPU | [Midas-V2.onnx](https://huggingface.co/qualcomm/Midas-V2/blob/main/Midas-V2.onnx) |
55
 
56
 
57
 
 
 
 
 
 
 
58
 
59
  ## Installation
60
 
 
110
  ```bash
111
  python -m qai_hub_models.models.midas.export
112
  ```
 
113
  ```
114
+ Profiling Results
115
+ ------------------------------------------------------------
116
+ Midas-V2
117
+ Device : Samsung Galaxy S23 (13)
118
+ Runtime : TFLITE
119
+ Estimated inference time (ms) : 3.2
120
+ Estimated peak memory usage (MB): [0, 2]
121
+ Total # Ops : 138
122
+ Compute Unit(s) : NPU (138 ops)
123
  ```
124
 
125
 
 
218
  Get more details on Midas-V2's performance across various devices [here](https://aihub.qualcomm.com/models/midas).
219
  Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/)
220
 
221
+
222
  ## License
223
+ * The license for the original implementation of Midas-V2 can be found [here](https://github.com/isl-org/MiDaS/blob/master/LICENSE).
224
+ * The license for the compiled assets for on-device deployment can be found [here](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf)
225
+
226
+
227
 
228
  ## References
229
  * [Towards Robust Monocular Depth Estimation: Mixing Datasets for Zero-shot Cross-dataset Transfer](https://arxiv.org/abs/1907.01341v3)
230
  * [Source Model Implementation](https://github.com/isl-org/MiDaS)
231
 
232
+
233
+
234
  ## Community
235
  * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
236
  * For questions or feedback please [reach out to us](mailto:[email protected]).