Examples demonstrating how to optimize caffe/tensorflow/darknet models with TensorRT and run inferencing on NVIDIA Jetson or x86_64 PC platforms. Highlights: (The following FPS numbers have been updated using test results against JetPack 4.3, i.e. TensorRT 6, on Jetson Nano.)
- Run an optimized 'GoogLeNet' image classifier at ~60 FPS on Jetson Nano.
- Run a very accurate optimized 'MTCNN' face detector at 6~11 FPS on Jetson Nano.
- Run an optimized 'ssd_mobilenet_v1_coco' object detector ('trt_ssd_async.py') at 27~28 FPS on Jetson Nano.
- Run an optimized 'yolov3-416' object detector at ~3 FPS on Jetson Nano.
- All demos work on Jetson TX2 and AGX Xavier (link and link), and run much faster!
- Furthermore, all demos should work on x86_64 PC with NVIDIA GPU(s) as well. Some minor tweaks would be needed. Please refer to README_x86.md for more information.
The code in this repository was tested on both Jetson Nano and Jetson TX2 Devkits. In order to run the demos below, first make sure you have the proper version of image (JetPack) installed on the target Jetson system. For example, reference for Jetson Nano: Setting up Jetson Nano: The Basics.
More specifically, the target Jetson system must have TensorRT libraries installed. Demo #1 and Demo #2 should work for TensorRT 3.x, 4.x, 5.x, 6.x. But Demo #3 and Demo #4 would require TensorRT 5.x or 6.x.
You could check which version of TensorRT has been installed on your Jetson system by looking at file names of the libraries. For example, TensorRT v5.1.6 (from JetPack-4.2.2) was present on my Jetson Nano DevKit.
$ ls /usr/lib/aarch64-linux-gnu/libnvinfer.so*
/usr/lib/aarch64-linux-gnu/libnvinfer.so
/usr/lib/aarch64-linux-gnu/libnvinfer.so.5
/usr/lib/aarch64-linux-gnu/libnvinfer.so.5.1.6Furthermore, the demo programs require 'cv2' (OpenCV) module in python3. You could, for example, refer to Installing OpenCV 3.4.6 on Jetson Nano for how to install opencv-3.4.6 on your Jetson system.
Lastly, if you plan to run Demo #3 (SSD), you'd also need to have 'tensorflowi-1.x' installed. You could refer to Building TensorFlow 1.12.2 on Jetson Nano for how to install tensorflow-1.12.2 on the Jetson system.
This demo illustrates how to convert a prototxt file and a caffemodel file into a tensorrt engine file, and to classify images with the optimized tensorrt engine.
Step-by-step:
-
Clone this repository.
$ cd ${HOME}/project $ git clone https://github.com/jkjung-avt/tensorrt_demos $ cd tensorrt_demos
-
Build the TensorRT engine from the trained googlenet (ILSVRC2012) model. Note that I downloaded the trained model files from BVLC caffe and have put a copy of all necessary files in this repository.
$ cd ${HOME}/project/tensorrt_demos/googlenet $ make $ ./create_engine
-
Build the Cython code. Install Cython if not previously installed.
$ pip3 install Cython $ cd ${HOME}/project/tensorrt_demos $ make
-
Run the 'trt_googlenet.py' demo program. For example, run the demo with a USB webcam as the input.
$ cd ${HOME}/project/tensorrt_demos $ python3 trt_googlenet.py --usb --vid 0 --width 1280 --height 720
Here's a screenshot of the demo (JetPack-4.2.2, i.e. TensorRT 5).
-
The demo program supports a number of different image inputs. You could do
python3 trt_googlenet.py --helpto read the help messages. Or more specifically, the following inputs could be specified:--file --filename test_video.mp4: a video file, e.g. mp4 or ts.--image --filename test_image.jpg: an image file, e.g. jpg or png.--usb --vid 0: USB webcam (/dev/video0).--rtsp --uri rtsp://admin:123456@192.168.1.1/live.sdp: RTSP source, e.g. an IP cam.
-
Check out my blog post for implementation details:
This demo builds upon the previous example. It converts 3 sets of prototxt and caffemodel files into 3 tensorrt engines, namely the PNet, RNet and ONet. Then it combines the 3 engine files to implement MTCNN, a very good face detector.
Assuming this repository has been cloned at '${HOME}/project/tensorrt_demos', follow these steps:
-
Build the TensorRT engines from the trained MTCNN model. (Refer to mtcnn/README.md for more information about the prototxt and caffemodel files.)
$ cd ${HOME}/project/tensorrt_demos/mtcnn $ make $ ./create_engines
-
Build the Cython code if it has not been done yet. Refer to step 3 in Demo #1.
-
Run the 'trt_mtcnn.py' demo program. For example, I just grabbed from the internet a poster of The Avengers for testing.
$ cd ${HOME}/project/tensorrt_demos $ python3 trt_mtcnn.py --image --filename ${HOME}/Pictures/avengers.jpg
Here's the result (JetPack-4.2.2, i.e. TensorRT 5).
-
The 'trt_mtcnn.py' demo program could also take various image inputs. Refer to step 5 in Demo #1 for details.
-
Check out my related blog posts:
This demo shows how to convert trained tensorflow Single-Shot Multibox Detector (SSD) models through UFF to TensorRT engines, and to do real-time object detection with the optimized TensorRT engines.
NOTE: This particular demo requires TensorRT 'Python API', which is only available in TensorRT 5.x+ on the Jetson systems. In other words, this demo only works on Jetson systems properly set up with JetPack-4.2+, but not JetPack-3.x or earlier versions.
Assuming this repository has been cloned at '${HOME}/project/tensorrt_demos', follow these steps:
-
Install requirements (pycuda, etc.) and build TensorRT engines from the trained SSD models.
$ cd ${HOME}/project/tensorrt_demos/ssd $ ./install.sh $ ./build_engines.sh
NOTE: On my Jetson Nano DevKit with TensorRT 5.1.6, the version number of UFF converter was "0.6.3". When I ran 'build_engine.py', the UFF library actually printed out:
UFF has been tested with tensorflow 1.12.0. Other versions are not guaranteed to work.So I would strongly suggest you to use tensorflow 1.12.x (or whatever matching version for the UFF library installed on your system) when converting pb to uff. -
Run the 'trt_ssd.py' demo program. The demo supports 4 models: 'ssd_mobilenet_v1_coco', 'ssd_mobilenet_v1_egohands', 'ssd_mobilenet_v2_coco', or 'ssd_mobilenet_v2_egohands'. For example, I tested the 'ssd_mobilenet_v1_coco' model with the 'huskies' picture.
$ cd ${HOME}/project/tensorrt_demos $ python3 trt_ssd.py --model ssd_mobilenet_v1_coco \ --image \ --filename ${HOME}/project/tf_trt_models/examples/detection/data/huskies.jpg
Here's the result (JetPack-4.2.2, i.e. TensorRT 5). Frame rate was good (over 20 FPS).
NOTE: When running this demo with TensorRT 6 (JetPack-4.3) on the Jetson Nano, I encountered the following error message which could probably be ignored for now. Quote from NVIDIA's NVES_R:
This is a known issue and will be fixed in a future version.[TensorRT] ERROR: Could not register plugin creator: FlattenConcat_TRT in namespaceI also tested the 'ssd_mobilenet_v1_egohands' (hand detector) model with a video clip from YouTube, and got the following result. Again, frame rate was pretty good. But the detection didn't seem very accurate though :-(
$ python3 trt_ssd.py --model ssd_mobilenet_v1_egohands \ --file \ --filename ${HOME}/Videos/Nonverbal_Communication.mp4(Click on the image below to see the whole video clip...)
-
The 'trt_ssd.py' demo program could also take various image inputs. Refer to step 5 in Demo #1 again.
-
Referring to this comment, '#TODO enable video pipeline', in the original TRT_object_detection code, I did implement an 'async' version of ssd detection code to do just that. When I tested 'ssd_mobilenet_v1_coco' on the same huskies image with the async demo program, frame rate improved 3~4 FPS.
$ cd ${HOME}/project/tensorrt_demos $ python3 trt_ssd_async.py --model ssd_mobilenet_v1_coco \ --image \ --filename ${HOME}/project/tf_trt_models/examples/detection/data/huskies.jpg
-
To verify accuracy (mAP) of the optimized TensorRT engines and make sure they do not degrade too much (due to reduced floating-point precision of 'FP16') from the original TensorFlow frozen inference graphs, you could prepare validation data and run 'eval_ssd.py'. Refer to README_eval_ssd.md for details.
I compared mAP of the TensorRT engine and the original tensorflow model for both 'ssd_mobilenet_v1_coco' and 'ssd_mobilenet_v2_coco' using COCO 'val2017' data. The results were good. In both cases, mAP of the optimized TensorRT engine matched the original tensorflow model. The FPS (frames per second) numbers in the table were measured using 'trt_ssd_async.py' on my Jetson Nano DevKit with JetPack-4.3.
TensorRT engine mAP @
IoU=0.5:0.95mAP @
IoU=0.5FPS on Nano mobilenet_v1 TF 0.232 0.351 -- mobilenet_v1 TRT (FP16) 0.232 0.351 27.7 mobilenet_v2 TF 0.248 0.375 -- mobilenet_v2 TRT (FP16) 0.248 0.375 22.7 -
Check out my blog posts for implementation details:
- TensorRT UFF SSD
- Speeding Up TensorRT UFF SSD
- Verifying mAP of TensorRT Optimized SSD and YOLOv3 Models
- Or if you'd like to learn how to train your own custom object detectors which could be easily converted to TensorRT engines and inferenced with 'trt_ssd.py' and 'trt_ssd_async.py': Training a Hand Detector with TensorFlow Object Detection API
Along the same line as Demo #3, this demo showcases how to convert a trained YOLOv3 model through ONNX to a TensorRT engine. This demo also requires TensorRT 'Python API' and has been verified working against both TensorRT 5.x and 6.x.
Assuming this repository has been cloned at '${HOME}/project/tensorrt_demos', follow these steps:
-
Install 'pycuda' in case you have not done so in Demo #3. Note that installation script resides in the 'ssd' folder.
$ cd ${HOME}/project/tensorrt_demos/ssd $ ./install_pycuda.sh
-
Install version '1.4.1' (not the latest) of python3 'onnx' module. Reference: information provided by NVIDIA.
$ sudo pip3 install onnx==1.4.1
-
Download the trained YOLOv3 COCO models and convert the targeted model to ONNX and then to TensorRT engine. This demo supports 5 models: 'yolov3-tiny-288', 'yolov3-tiny-416', 'yolov3-288', 'yolov3-416', and 'yolov3-608'. NOTE: I'm not sure whether my implementation of the 'yolov3-tiny-288' and 'yolov3-tiny-416' models is correct. They are for reference only.
I use 'yolov3-416' as example below.
$ cd ${HOME}/project/tensorrt_demos/yolov3_onnx $ ./download_yolov3.sh $ python3 yolov3_to_onnx.py --model yolov3-416 $ python3 onnx_to_tensorrt.py --model yolov3-416
The last step ('onnx_to_tensorrt.py') takes a little bit more than half au hour to complete on my Jetson Nano DevKit. When that is done, the optimized TensorRT engine would be saved as 'yolov3-416.trt'.
-
Test the YOLOv3 TensorRT engine with the 'dog.jpg' image.
$ wget https://raw.githubusercontent.com/pjreddie/darknet/master/data/dog.jpg -O ${HOME}/Pictures/dog.jpg $ python3 trt_yolov3.py --model yolov3-416 --image --filename ${HOME}/Pictures/dog.jpg
This was tested against JetPack-4.3, i.e. TensorRT 6.
-
The 'trt_yolov3.py' demo program could also take various image inputs. Refer to step 5 in Demo #1 again.
-
I created 'eval_yolov3.py' for evaluating mAP of the optimized YOLOv3 engine. It works the same way as 'eval_ssd.py'. Refer to step #5 in Demo #3.
$ python3 eval_yolov3.py --model yolov3-288 $ python3 eval_yolov3.py --model yolov3-416 $ python3 eval_yolov3.py --model yolov3-608
I evaluated all of yolov3-tiny-288, yolov3-tiny-416, yolov3-288, yolov3-416 and yolov3-608 TensorRT engines with COCO 'val2017' data and got the following results. The FPS (frames per second) numbers were measured using 'trt_yolov3.py' on my Jetson Nano DevKit with JetPack-4.3.
TensorRT engine mAP @
IoU=0.5:0.95mAP @
IoU=0.5FPS on Nano yolov3-tiny-288 (FP16) 0.077 0.158 20.9 yolov3-tiny-416 (FP16) 0.096 0.202 14.2 yolov3-288 (FP16) 0.331 0.600 5.42 yolov3-416 (FP16) 0.373 0.664 3.07 yolov3-608 (FP16) 0.376 0.665 1.53 yolov3-608 (FP32) 0.376 0.665 -- -
Check out my blog post for implementation details:
I referenced source code of NVIDIA/TensorRT samples to develop most of the demos in this repository. Those NVIDIA samples are under Apache License 2.0.