Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,79 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
pipeline_tag: depth-estimation
|
| 4 |
+
---
|
| 5 |
+
# FastDepth
|
| 6 |
+
|
| 7 |
+
## **Use case** : `Depth Estimation`
|
| 8 |
+
|
| 9 |
+
# Model description
|
| 10 |
+
|
| 11 |
+
FastDepth is a lightweight encoder-decoder network designed for real-time monocular depth estimation, optimized for edge devices. This implementation is based on model number 146 from [PINTO's model zoo](https://github.com/PINTO0309/PINTO_model_zoo), which builds upon a MobileNetV1 based feature extractor and a fast decoder.
|
| 12 |
+
|
| 13 |
+
Although the original training dataset is not explicitly provided, it is most likely **NYU Depth V2**, a standard benchmark dataset for indoor depth estimation.
|
| 14 |
+
|
| 15 |
+
## Network information
|
| 16 |
+
|
| 17 |
+
|
| 18 |
+
| Network Information | Value |
|
| 19 |
+
|-------------------------|----------------------------------------------------------------|
|
| 20 |
+
| Framework | TensorFlowLite |
|
| 21 |
+
| Quantization | int8 |
|
| 22 |
+
| Provenance | PINTO Model Zoo #146 |
|
| 23 |
+
| Paper | [Link to Paper](https://arxiv.org/pdf/1903.03273)|
|
| 24 |
+
|
| 25 |
+
The models are quantized using tensorflow lite converter.
|
| 26 |
+
|
| 27 |
+
|
| 28 |
+
## Network inputs / outputs
|
| 29 |
+
|
| 30 |
+
| Input Shape | Description |
|
| 31 |
+
|--------------|-----------------------------------------------------|
|
| 32 |
+
| (1, H, W, 3) | Single RGB image (int8) |
|
| 33 |
+
|
| 34 |
+
| Output Shape | Description |
|
| 35 |
+
|---------------|-------------------------------------------------|
|
| 36 |
+
| (1, H, W, 1) | Single-channel depth prediction (int8)|
|
| 37 |
+
|
| 38 |
+
|
| 39 |
+
## Recommended platforms
|
| 40 |
+
|
| 41 |
+
|
| 42 |
+
| Platform | Supported | Recommended |
|
| 43 |
+
|----------|--------|-----------|
|
| 44 |
+
| STM32L0 |[]|[]|
|
| 45 |
+
| STM32L4 |[]|[]|
|
| 46 |
+
| STM32U5 |[]|[]|
|
| 47 |
+
| STM32H7 |[]|[]|
|
| 48 |
+
| STM32MP1 |[]|[]|
|
| 49 |
+
| STM32MP2 |[x]|[x]|
|
| 50 |
+
| STM32N6 |[x]|[x]|
|
| 51 |
+
|
| 52 |
+
|
| 53 |
+
# Performances
|
| 54 |
+
|
| 55 |
+
## Metrics
|
| 56 |
+
|
| 57 |
+
Measures are done with default STM32Cube.AI configuration with enabled input / output allocated option.
|
| 58 |
+
|
| 59 |
+
### Reference **NPU** memory footprint
|
| 60 |
+
|
| 61 |
+
| Model | Dataset | Format | Resolution | Series | Internal RAM (KiB) | External RAM (KiB) | Weights Flash (KiB) | STM32Cube.AI version | STEdgeAI Core version |
|
| 62 |
+
|------------|---------------|----------|------------|-----------|--------------|--------------|---------------|----------------------|-----------------------|
|
| 63 |
+
| [Fast Depth](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/pose_estimation/fast_depth/Public_pretrainedmodel_public_dataset/nyu_depth_v2/fast_depth_224/fast_depth_224_int8_pc.tflite) | NYU depth v2 | Int8 | 224x224x3 | STM32N6 | 2365.98 | 0.0 | 1505.19 | 10.2.0 | 2.2.0 |
|
| 64 |
+
| [Fast Depth](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/pose_estimation/fast_depth/Public_pretrainedmodel_public_dataset/nyu_depth_v2/fast_depth_224/fast_depth_224_int8_pc.tflite) | NYU depth v2 | Int8 | 256x256x3 | STM32N6 | 2688 | 1024.0 | 1505.19 | 10.2.0 | 2.2.0 |
|
| 65 |
+
| [Fast Depth](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/pose_estimation/fast_depth/Public_pretrainedmodel_public_dataset/nyu_depth_v2/fast_depth_320/fast_depth_320_int8_pc.tflite) | NYU depth v2 | Int8 | 224x224x3 | STM32N6 | 2800 | 1600 | 1505.17 | 10.2.0 | 2.2.0 |
|
| 66 |
+
|
| 67 |
+
|
| 68 |
+
### Reference **NPU** inference time
|
| 69 |
+
|
| 70 |
+
|
| 71 |
+
| Model | Dataset | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STM32Cube.AI version | STEdgeAI Core version |
|
| 72 |
+
|------------|---------------|----------|------------|------------------|------------------|---------------------|-------------|----------------------|-------------------------|
|
| 73 |
+
| [Fast Depth](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/pose_estimation/fast_depth/Public_pretrainedmodel_public_dataset/nyu_depth_v2/fast_depth_224/fast_depth_224_int8_pc.tflite) | NYU depth v2 | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 24.43 | 40.93 | 10.2.0 | 2.2.0 |
|
| 74 |
+
| [Fast Depth](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/pose_estimation/fast_depth/Public_pretrainedmodel_public_dataset/nyu_depth_v2/fast_depth_224/fast_depth_224_int8_pc.tflite) | NYU depth v2 | Int8 | 256x256x3 | STM32N6570-DK | NPU/MCU | 55.51 | 18.01 | 10.2.0 | 2.2.0 |
|
| 75 |
+
| [Fast Depth](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/pose_estimation/fast_depth/Public_pretrainedmodel_public_dataset/nyu_depth_v2/fast_depth_320/fast_depth_320_int8_pc.tflite) | NYU depth v2 | Int8 | 320x320x3 | STM32N6570-DK | NPU/MCU | 56.07 | 17.83 | 10.2.0 | 2.2.0 |
|
| 76 |
+
|
| 77 |
+
|
| 78 |
+
|
| 79 |
+
Please refer to the stm32ai-modelzoo-services GitHub [here](https://github.com/STMicroelectronics/stm32ai-modelzoo-services)
|