|
--- |
|
license: apache-2.0 |
|
--- |
|
<img src="https://cdn-uploads.huggingface.co/production/uploads/64c1fef5b9d81735a12c3fcc/sebNQgVO1hUapijWvVwTl.jpeg" width=600> |
|
|
|
# YOLOv5: Target Detection |
|
|
|
Yolov5 is a one-stage structure target detection network framework, in which the main structure consists of 4 parts, including the network backbone composed of modified CSPNet, the high-resolution feature fusion module composed of FPN (Feature Paramid Network), composed of SPP (Spatial Pyramid Pooling) constitutes a pooling module, and three different detection heads are used to detect targets of different sizes. |
|
|
|
The YOLOv5 model can be found [here](https://github.com/ultralytics/yolov5) |
|
|
|
## CONTENTS |
|
- [Source Model](#source-model) |
|
- [Performance](#performance) |
|
- [Model Conversion](#model-conversion) |
|
- [Tutorial](#tutorial) |
|
|
|
## Source Model |
|
|
|
The steps followed the [yolov5 tutorials](https://docs.ultralytics.com/yolov5/tutorials/model_export/) to get the source model in ONNX format. |
|
|
|
> The source model **YOLOv5s.onnx** also can be found [here](https://huggingface.co/aplux/YOLOv5/blob/main/yolov5s.onnx). |
|
|
|
**Environment Preparation** |
|
```bash |
|
git clone https://github.com/ultralytics/yolov5 # clone |
|
cd yolov5 |
|
pip install -r requirements.txt # install |
|
``` |
|
|
|
**Export to ONNX** |
|
|
|
```bash |
|
python export.py --weights yolov5s.pt --include torchscript onnx --opset 12 |
|
``` |
|
|
|
## Performance |
|
|
|
<center><b>🧰QCS6490</b></center> |
|
|
|
|Device|Runtime|Model|Size (pixels)|Inference Time (ms)|Precision|Compute Unit|Model Download| |
|
|:----:|:----:|:----:|:----:|:----:|:----:|:----:|:----:| |
|
|AidBox QCS6490|QNN|YOLOv5s(cutoff)|640|6.7|INT8|NPU|[model download](https://huggingface.co/aidlux/YOLOv5/blob/main/models/QCS6490/cutoff_yolov5s_int8_qnn/cutoff_yolov5s_int8.qnn.serialized.bin)| |
|
|AidBox QCS6490|QNN|YOLOv5s(cutoff)|640|15.2|INT16|NPU|[model download](https://huggingface.co/aidlux/YOLOv5/blob/main/models/QCS6490/cutoff_yolov5s_int16_qnn/cutoff_yolov5s_int16.qnn.serialized.bin)| |
|
|AidBox QCS6490|SNPE|YOLOv5s(cutoff)|640|5.5|INT8|NPU|[model download](https://huggingface.co/aidlux/YOLOv5/blob/main/models/QCS6490/cutoff_yolov5s_int8_htp_snpe2/cutoff_yolov5s_int8_htp_snpe2.dlc)| |
|
|AidBox QCS6490|SNPE|YOLOv5s(cutoff)|640|13.4|INT16|NPU|[model download](https://huggingface.co/aidlux/YOLOv5/blob/main/models/QCS6490/cutoff_yolov5s_int16_htp_snpe2/cutoff_yolov5s_int16_htp_snpe2.dlc)| |
|
|
|
<center><b>🧰QCS8550</b></center> |
|
|
|
|Device|Runtime|Model|Size (pixels)|Inference Time (ms)|Precision|Compute Unit|Model Download| |
|
|:----:|:----:|:----:|:----:|:----:|:----:|:----:|:----:| |
|
|APLUX QCS8550|QNN|YOLOv5s(cutoff)|640|4.1|INT8|NPU|[model download](https://huggingface.co/aidlux/YOLOv5/blob/main/models/QCS8550/cutoff_yolov5s_int8_qnn/cutoff_yolov5s_640_int8.qnn.serialized.bin)| |
|
|APLUX QCS8550|QNN|YOLOv5s(cutoff)|640|13.4|INT16|NPU|[model download](https://huggingface.co/aidlux/YOLOv5/blob/main/models/QCS8550/cutoff_yolov5s_int16_qnn/cutoff_yolov5s_640_int16.qnn.serialized.bin)| |
|
|APLUX QCS8550|SNPE|YOLOv5s(cutoff)|640|2.3|INT8|NPU|[model download](https://huggingface.co/aidlux/YOLOv5/blob/main/models/QCS8550/cutoff_yolov5s_int8_htp_snpe2/cutoff_yolov5s_int8_htp_snpe2.dlc)| |
|
|APLUX QCS8550|SNPE|YOLOv5s(cutoff)|640|5.8|INT16|NPU|[model download](https://huggingface.co/aidlux/YOLOv5/blob/main/models/QCS8550/cutoff_yolov5s_int16_htp_snpe2/cutoff_yolov5s_int16_htp_snpe2.dlc)| |
|
|
|
## Model Conversion |
|
|
|
Demo models converted from [**AIMO(AI Model Optimizier)**](https://aidlux.com/en/product/aimo). |
|
|
|
The demo model conversion step on AIMO can be found blow: |
|
|
|
<center><b>🧰QCS6490</b></center> |
|
|
|
|Device|Runtime|Model|Size (pixels)|Precision|Compute Unit|AIMO Conversion Steps| |
|
|:----:|:----:|:----:|:----:|:----:|:----:|:----:| |
|
|AidBox QCS6490|QNN|YOLOv5s(cutoff)|640|INT8|NPU|[view steps](https://huggingface.co/aplux/YOLOv5/blob/main/aimo/QCS6490/aimo_yolov5s_qnn_int8.png)| |
|
|AidBox QCS6490|QNN|YOLOv5s(cutoff)|640|INT16|NPU|[view steps](https://huggingface.co/aplux/YOLOv5/blob/main/aimo/QCS6490/aimo_yolov5s_qnn_int16.png)| |
|
|AidBox QCS6490|SNPE|YOLOv5s(cutoff)|640|INT8|NPU|[view steps](https://huggingface.co/aplux/YOLOv5/blob/main/aimo/QCS6490/aimo_yolov5s_snpe_int8.png)| |
|
|AidBox QCS6490|SNPE|YOLOv5s(cutoff)|640|INT16|NPU|[view steps](https://huggingface.co/aplux/YOLOv5/blob/main/aimo/QCS6490/aimo_yolov5s_snpe_int16.png)| |
|
|
|
<center><b>🧰QCS8550</b></center> |
|
|
|
|Device|Runtime|Model|Size (pixels)|Precision|Compute Unit|AIMO Conversion Steps| |
|
|:----:|:----:|:----:|:----:|:----:|:----:|:----:| |
|
|APLUX QCS8550|QNN|YOLOv5s(cutoff)|640|INT8|NPU|[view steps](https://huggingface.co/aplux/YOLOv5/blob/main/aimo/QCS8550/aimo_yolov5s_qnn_int8.png)| |
|
|APLUX QCS8550|QNN|YOLOv5s(cutoff)|640|INT16|NPU|[view steps](https://huggingface.co/aplux/YOLOv5/blob/main/aimo/QCS8550/aimo_yolov5s_qnn_int16.png)| |
|
|APLUX QCS8550|SNPE|YOLOv5s(cutoff)|640|INT8|NPU|[view steps](https://huggingface.co/aplux/YOLOv5/blob/main/aimo/QCS8550/aimo_yolov5s_snpe_int8.png)| |
|
|APLUX QCS8550|SNPE|YOLOv5s(cutoff)|640|INT16|NPU|[view steps](https://huggingface.co/aplux/YOLOv5/blob/main/aimo/QCS8550/aimo_yolov5s_snpe_int16.png)| |
|
|
|
## Tutorial |
|
|
|
### Step1: convert model |
|
|
|
1.1 Prepare source model in onnx format. The source model can be found [here](https://huggingface.co/aplux/YOLOv5/blob/main/yolov5s.onnx) or following [Source Model](#source-model) to obtain. |
|
|
|
1.2 Login [AIMO](https://aidlux.com/en/product/aimo) and convert source model to target format. The model conversion step can follow **AIMO Conversion Step** in [Model Conversion Sheet](#model-conversion). |
|
|
|
1.3 After conversion task done, download target model file. |
|
|
|
> note: you can skip convert model step, and directly download converted model in [Performance Sheet](#performance). |
|
|
|
### Step2: install AidLite SDK |
|
|
|
```bash |
|
# install aidlite sdk c++ api |
|
sudo aid-pkg -i aidlite-sdk |
|
# install aidlite sdk python api |
|
python3 -m pip install pyaidlite -i https://mirrors.aidlux.com --trusted-host mirrors.aidlux.com |
|
``` |
|
|
|
The developer document of AidLite SDK can be found [here](https://huggingface.co/datasets/aplux/AIToolKit/blob/main/AidLite%20SDK%20Development%20Documents.md). |
|
|
|
### Step3: model inference |
|
|
|
3.1 Download demo program |
|
```bash |
|
# download demo program |
|
wget https://huggingface.co/aplux/YOLOv5/resolve/main/examples.zip |
|
# unzip |
|
unzip examples.zip |
|
``` |
|
|
|
3.2 Modify `model_path` to your model path and run demo |
|
|
|
```bash |
|
# run qnn demo |
|
python qnn_yolov5_multi.py |
|
# run snpe demo |
|
python snpe2_yolov5_multi.py |
|
``` |
