Davide Sapienza
209 lines
13 KiB
Markdown
209 lines
13 KiB
Markdown
# tkDNN
|
|
tkDNN is a Deep Neural Network library built with cuDNN and tensorRT primitives, specifically thought to work on NVIDIA Embedded Boards. It has been tested on TK1, TX1, TX2, AGX Xavier and several discrete GPU.
|
|
The main goal of this project is to exploit NVIDIA boards as much as possible to obtain the best inference performance. It does not allow training.
|
|
|
|
## Index
|
|
* [Dependencies](#dependencies)
|
|
* [How to compile this repo](#how-to-compile-this-repo)
|
|
* [Workflow](#workflow)
|
|
* [How to export weights](#how-to-export-weights)
|
|
* [Run the demo](#run-the-demo)
|
|
* [mAP demo](#map-demo)
|
|
* [Existing tests and supported networks](#existing-tests-and-supported-networks)
|
|
* [References](#references)
|
|
|
|
|
|
|
|
|
|
## Dependencies
|
|
This branch works on every NVIDIA GPU that supports the dependencies:
|
|
* CUDA 10.0
|
|
* CUDNN 7.603
|
|
* TENSORRT 6.01
|
|
* OPENCV 4.1
|
|
* yaml-cpp 0.5.2 (sudo apt install libyaml-cpp-dev)
|
|
|
|
## How to compile this repo
|
|
Build with cmake. If using Ubuntu 18.04 a new version of cmake is needed (1.15 or above).
|
|
```
|
|
git clone https://github.com/ceccocats/tkDNN
|
|
git cd tkDNN
|
|
git checkout cnet
|
|
mkdir build
|
|
cd build
|
|
cmake .. # use -DTEST_DATA=False to skip dataset download
|
|
make
|
|
```
|
|
If TEST_DATA is not set to False, weights needed to run some tests will be automatically downloaded.
|
|
|
|
## Workflow
|
|
Steps needed to do inference on tkDNN with a custom neural network.
|
|
* Build and train a NN model with your favorite framework.
|
|
* Export weights and bias for each layer and save them in a binary file (one for layer).
|
|
* Export outputs for each layer and save them in a binary file (one for layer).
|
|
* Create a new test and define the network, layer by layer using the weights extracted and the output to check the results.
|
|
* Do inference.
|
|
|
|
## How to export weights
|
|
|
|
Weights are essential for any network to run inference. For each test a folder organized as follow is needed:
|
|
```
|
|
test_nn
|
|
|---- test_nn.cpp (nn definition in tkDNN)
|
|
|---- layers/ (folder containing a binary file for each layer with the corresponding wieghts and bias)
|
|
|---- debug/ (folder containing a binary file for each layer with the corresponding outputs)
|
|
```
|
|
Therefore, once the weights have been exported, the folders layers and debug should be placed in the corresponding test.
|
|
|
|
### 1)Export weights from darknet
|
|
To export weights for NNs that are defined in darknet framework, use [this](https://github.com/ceccocats/darknet) fork of darknet and follow these steps to obtain a correct debug and layers folder, ready for tkDNN.
|
|
|
|
```
|
|
git clone https://github.com/ceccocats/darknet
|
|
git cd darknet
|
|
make
|
|
mkdir layers debug
|
|
./darknet export <path-to-cfg-file> <path-to-weights> layers
|
|
```
|
|
N.b. Use compilation with CPU (leave GPU=0 in Makefile) if you also want debug.
|
|
|
|
### 2)Export weights for DLA34 and ResNet101
|
|
To get weights and outputs needed to run the tests dla34 and resnet101 use the Python script and the Anaconda environment included in the repository.
|
|
|
|
Create Anaconda environment and activate it:
|
|
```
|
|
conda env create -f file_name.yml
|
|
source activate env_name
|
|
python <script name>
|
|
```
|
|
### 3)Export weights for CenterNet
|
|
To get the weights needed to run Centernet tests use [this](https://github.com/sapienzadavide/CenterNet.git) fork of the original Centernet.
|
|
```
|
|
git clone https://github.com/sapienzadavide/CenterNet.git
|
|
```
|
|
* follow the instruction in the README.md and INSTALL.md
|
|
|
|
```
|
|
python demo.py --input_res 512 --arch resdcn_101 ctdet --demo /path/to/image/or/folder/or/video/or/webcam --load_model ../models/ctdet_coco_resdcn101.pth --exp_wo --exp_wo_dim 512
|
|
python demo.py --input_res 512 --arch dla_34 ctdet --demo /path/to/image/or/folder/or/video/or/webcam --load_model ../models/ctdet_coco_dla_2x.pth --exp_wo --exp_wo_dim 512
|
|
```
|
|
### 4)Export weights for MobileNetSSD
|
|
|
|
To get the weights needed to run Mobilenet tests use [this](https://github.com/mive93/pytorch-ssd) fork of a Pytorch implementation of SSD network.
|
|
|
|
```
|
|
git clone https://github.com/mive93/pytorch-ssd
|
|
cd pytorch-ssd
|
|
conda env create -f env_mobv2ssd.yml
|
|
python run_ssd_live_demo.py mb2-ssd-lite <pth-model-fil> <labels-file>
|
|
```
|
|
## Run the demo
|
|
|
|
To run the an object detection demo follow these steps (example with yolov3):
|
|
```
|
|
rm yolo3_FP32.rt # be sure to delete(or move) old tensorRT files
|
|
./test_yolo3 # run the yolo test (is slow)
|
|
./demo yolo3_FP32.rt ../demo/yolo_test.mp4 y
|
|
```
|
|
In general the demo program takes 3 parameters:
|
|
```
|
|
./demo <network-rt-file> <path-to-video> <kind-of-network>
|
|
```
|
|
where
|
|
* ```<network-rt-file>``` is the rt file generated by a test
|
|
* ```<<path-to-video>``` is the path to a video file or a camera input
|
|
* ```<kind-of-network>``` is the type of network. Thee types are currently supported: ```y``` (YOLO family), ```c``` (CenterNet family) and ```m``` (MobileNet-SSD family)
|
|
N.b. By default it is used FP32 inference
|
|
|
|
|
|
|
|
### FP16 inference
|
|
|
|
To run the an object detection demo with FP16 inference follow these steps (example with yolov3):
|
|
```
|
|
export TKDNN_MODE=FP16 # set the half floating point optimization
|
|
rm yolo3_FP16.rt # be sure to delete(or move) old tensorRT files
|
|
./test_yolo3 # run the yolo test (is slow)
|
|
./demo yolo3_FP16.rt ../demo/yolo_test.mp4 y
|
|
```
|
|
N.b. Using FP16 inference will lead to some errors in the results (first or second decimal).
|
|
|
|
### INT8 inference
|
|
|
|
To run the an object detection demo with INT8 inference follow these steps (example with yolov3):
|
|
```
|
|
export TKDNN_MODE=INT8 # set the 8-bit integer optimization
|
|
|
|
# image_list.txt contains the list of the absolute paths to the calibration images
|
|
export TKDNN_CALIB_IMG_PATH=/path/to/calibration/image_list.txt
|
|
|
|
# label_list.txt contains the list of the absolute paths to the calibration labels
|
|
export TKDNN_CALIB_LABEL_PATH=/path/to/calibration/label_list.txt
|
|
rm yolo3_INT8.rt # be sure to delete(or move) old tensorRT files
|
|
./test_yolo3 # run the yolo test (is slow)
|
|
./demo yolo3_INT8.rt ../demo/yolo_test.mp4 y
|
|
```
|
|
N.b. Using INT8 inference will lead to some errors in the results (first or second decimal).
|
|
|
|
N.b. The network test script is very slow: it call the INT8 calibrator.
|
|
|
|
## mAP demo
|
|
|
|
To compute mAP, precision, recall and f1score, run the map_demo.
|
|
|
|
A validation set is needed. To download COCO_val2017 run (form the root folder):
|
|
```
|
|
bash download_validation.sh
|
|
```
|
|
To compute the map, the following parameters are needed:
|
|
```
|
|
./map_demo <network rt> <network type [y|c|m]> <labels file path> <config file path>
|
|
```
|
|
where
|
|
* ```<network rt>```: rt file of a chosen network on which compute the mAP.
|
|
* ```<network type [y|c|m]>```: type of network. Right now only y(yolo), c(centernet) and m(mobilenet) are allowed
|
|
* ```<labels file path>```: path to a text file containing all the paths of the ground-truth labels. It is important that all the labels of the ground-truth are in a folder called 'labels'. In the folder containing the folder 'labels' there should be also a folder 'images', containing all the ground-truth images having the same same as the labels. To better understand, if there is a label path/to/labels/000001.txt there should be a corresponding image path/to/images/000001.jpg.
|
|
* ```<config file path>```: path to a yaml file with the parameters needed for the mAP computation, similar to demo/config.yaml
|
|
|
|
Example:
|
|
|
|
```
|
|
cd build
|
|
./map_demo dla34_cnet_FP32.rt c ../demo/COCO_val2017/all_labels.txt ../demo/config.yaml
|
|
```
|
|
|
|
## Existing tests and supported networks
|
|
|
|
| Test Name | Network | Dataset | N Classes | Input size | Weights |
|
|
| :---------------- | :-------------------------------------------- | :-----------------------------------------------------------: | :-------: | :-----------: | :------------------------------------------------------------------------ |
|
|
| yolo | YOLO v2<sup>1</sup> | [COCO 2014](http://cocodataset.org/) | 80 | 608x608 | [weights](https://cloud.hipert.unimore.it/s/nf4PJ3k8bxBETwL/download) |
|
|
| yolo_224 | YOLO v2<sup>1</sup> | [COCO 2014](http://cocodataset.org/) | 80 | 224x224 | weights |
|
|
| yolo_berkeley | YOLO v2<sup>1</sup> | [BDD100K ](https://bair.berkeley.edu/blog/2018/05/30/bdd/) | 10 | 416x736 | weights |
|
|
| yolo_relu | YOLO v2 (with ReLU, not Leaky)<sup>1</sup> | [COCO 2014](http://cocodataset.org/) | 80 | 416x416 | weights |
|
|
| yolo_tiny | YOLO v2 tiny<sup>1</sup> | [COCO 2014](http://cocodataset.org/) | 80 | 416x416 | [weights](https://cloud.hipert.unimore.it/s/m3orfJr8pGrN5mQ/download) |
|
|
| yolo_voc | YOLO v2<sup>1</sup> | [VOC ](http://host.robots.ox.ac.uk/pascal/VOC/) | 21 | 416x416 | [weights](https://cloud.hipert.unimore.it/s/DJC5Fi2pEjfNDP9/download) |
|
|
| yolo3 | YOLO v3<sup>2</sup> | [COCO 2014](http://cocodataset.org/) | 80 | 416x416 | [weights](https://cloud.hipert.unimore.it/s/jPXmHyptpLoNdNR/download) |
|
|
| yolo3_berkeley | YOLO v3<sup>2</sup> | [BDD100K ](https://bair.berkeley.edu/blog/2018/05/30/bdd/) | 10 | 320x544 | [weights](https://cloud.hipert.unimore.it/s/o5cHa4AjTKS64oD/download) |
|
|
| yolo3_coco4 | YOLO v3<sup>2</sup> | [COCO 2014](http://cocodataset.org/) | 4 | 416x416 | [weights](https://cloud.hipert.unimore.it/s/o27NDzSAartbyc4/download) |
|
|
| yolo3_flir | YOLO v3<sup>2</sup> | [FREE FLIR](https://www.flir.com/oem/adas/adas-dataset-form/) | 3 | 320x544 | [weights](https://cloud.hipert.unimore.it/s/62DECncmF6bMMiH/download) |
|
|
| yolo3_tiny | YOLO v3 tiny<sup>2</sup> | [COCO 2014](http://cocodataset.org/) | 80 | 416x416 | [weights](https://cloud.hipert.unimore.it/s/LMcSHtWaLeps8yN/download) |
|
|
| yolo3_tiny512 | YOLO v3 tiny<sup>2</sup> | [COCO 2017](http://cocodataset.org/) | 80 | 512x512 | [weights](https://cloud.hipert.unimore.it/s/njnYACnQfWQFKrn/download) |
|
|
| dla34 | Deep Leayer Aggreagtion (DLA) 34<sup>3</sup> | [COCO 2014](http://cocodataset.org/) | 80 | 224x224 | weights |
|
|
| dla34_cnet | Centernet (DLA34 backend)<sup>4</sup> | [COCO 2017](http://cocodataset.org/) | 80 | 512x512 | [weights](https://cloud.hipert.unimore.it/s/8AjXdgCeRzCa5AF/download) |
|
|
| mobilenetv2ssd | Mobilnet v2 SSD Lite<sup>5</sup> | [VOC ](http://host.robots.ox.ac.uk/pascal/VOC/) | 21 | 300x300 | [weights](https://cloud.hipert.unimore.it/s/x4ZfxBKN23zAJQp/download) |
|
|
| resnet101 | Resnet 101<sup>6</sup> | [COCO 2014](http://cocodataset.org/) | 80 | 224x224 | weights |
|
|
| resnet101_cnet | Centernet (Resnet101 backend)<sup>4</sup> | [COCO 2017](http://cocodataset.org/) | 80 | 512x512 | [weights](https://cloud.hipert.unimore.it/s/B6mj33k7beECXsY/download) |
|
|
| csresnext50-panet-spp | Cross Stage Partial Network <sup>7</sup> | [COCO 2014](http://cocodataset.org/) | 80 | 416x416 | [weights](https://cloud.hipert.unimore.it/s/Kcs4xBozwY4wFx8/download) |
|
|
|
|
|
|
|
|
## References
|
|
|
|
1. Redmon, Joseph, and Ali Farhadi. "YOLO9000: better, faster, stronger." Proceedings of the IEEE conference on computer vision and pattern recognition. 2017.
|
|
2. Redmon, Joseph, and Ali Farhadi. "Yolov3: An incremental improvement." arXiv preprint arXiv:1804.02767 (2018).
|
|
3. Yu, Fisher, et al. "Deep layer aggregation." Proceedings of the IEEE conference on computer vision and pattern recognition. 2018.
|
|
4. Zhou, Xingyi, Dequan Wang, and Philipp Krähenbühl. "Objects as points." arXiv preprint arXiv:1904.07850 (2019).
|
|
5. Sandler, Mark, et al. "Mobilenetv2: Inverted residuals and linear bottlenecks." Proceedings of the IEEE conference on computer vision and pattern recognition. 2018.
|
|
6. He, Kaiming, et al. "Deep residual learning for image recognition." Proceedings of the IEEE conference on computer vision and pattern recognition. 2016.
|
|
7. Wang, Chien-Yao, et al. "CSPNet: A New Backbone that can Enhance Learning Capability of CNN." arXiv preprint arXiv:1911.11929 (2019).
|