Tiny-YOLO-v2 ROS Node for Traffic Light Detection

Original By @pjreddie, @thtrieu and @udacity

Current By: Chris Gundling, chrisgundling@gmail.com

---

Darkflow was originally implemented by @thtrieu. The original Tiny-YOLOv2 model was then trained against the annotated Udacity SDC datasets by several @udacity open source community members. The model that they trained is capable of detecting cars, trucks, bicycles, pedestrians and traffic lights. This YOLOv2 model has been modified to be a traffic light detector and was implemented as a ROS node that should be capable of real time operation in ROS on a high performance GPU. The package runs at 23Hz on a NVIDIA GTX 860M. (/scripts/times.txt currently shows benchmarking stats on Hz using an AWS K520, but will update once I can do more high performance GPU testing). Here is an image of the model runnning standalone outside of ROS and a youtube video.

ROS Node

This repository can be used as a ROS package. It includes the python class TLightNode inside of /scripts/tlight_node.py that is run with /scripts/darkflow.py. The traffic light detector node will subscribe to an Image topic and the publish the same image with traffic lights detected using boxes and annotations. A screenshot of the ROS node operating within ROS is shown below, with original raw image on top and published image below.

*Note that I only have ROS installed on a CPU at the moment, so there is a significant lag in the published images.

Background Info

Real-time object detection and classification. Paper: version 1, version 2.

Read more about YOLO (in darknet) and download the original weight files here. In case the weight file cannot be found, @thtrieu also uploaded some of his here, which include yolo-full and yolo-tiny of v1.0, tiny-yolo-v1.1 of v1.1 and yolo, tiny-yolo-voc of v2.

Training Against Udacity Self Driving Datasets

Udacity Self Driving Car Open Source Project have provided an annotated dataset of images that contains bounding boxes for five classes of objects: cars, pedestrians, truck, cyclists and traffic lights.

The v2 tiny-yolo configuration that was used for this ROS node, trained on the udacity dataset can be found at cfg/tiny-yolo-udacity.cfg, with checkpoint here. To run this this package you will need to download this checkpoint and put files in a /scripts/ckpt/ directory.

Dependencies

Python 2.7, tensorflow 1.0, numpy, opencv, ROS 1.0.

Getting Things Running

Create a ROS package called yolo_light using catkin_create_pkg yolo_light. Clone this repository and copy all contents into the yolo_light package. You should then perform the following operations.

1. Build the Cython extensions in place.

   python setup.py build_ext --inplace
   

2. You then have two options, for testing, you can run the detector standalone from ROS using the following command:

   python run.py
   

   or, after performing a ROS catkin_make, you can start the node from the /yolo_light/scripts directory using:

   rosrun yolo_light darkflow.py
   

Running Stadalone (Outside ROS) darkflow with run.py

When performing a python run.py, due to the difference in images that are read in from ROS, two changes within the code must be made.

On line 17-18 in yolo_light/scripts/net/yolo/test.py you must uncomment line 17 and comment line 18.

On line 73-74 in yolo_light/scripts/net/flow.py you must comment line 73 and uncomment line 74.

In return_predict function in yolo_light/scripts/net/flow.py add the input imname.

More Info About Other Features of Darkflow

Seperate from what is decribed above, for testing outside of ROS, you can also just run flow.

# Have a look at its options
./flow --h

All input images from default folder test/ are flowed through the net and predictions are put in test/out/. We can always specify more parameters for such forward passes, such as detection threshold, batch size, test folder, etc.

I used the following command for the current v2 tiny-yolo model:

./flow --test test/ --model cfg/tiny-yolo-udacity.cfg --load 8987 --gpu 1.0

For some other functionality, consider the following:

# 1. Load yolo-tiny.weights
./flow --model cfg/yolo-tiny.cfg --load bin/yolo-tiny.weights

# 2. To completely initialize a model, leave the --load option
./flow --model cfg/yolo-3c.cfg

# 3. It is useful to reuse the first identical layers of tiny for 3c
./flow --model cfg/yolo-3c.cfg --load bin/yolo-tiny.weights
# this will print out which layers are reused, which are initialized

json output can be generated with descriptions of the pixel location of each bounding box and the pixel location. Each prediction is stored in the test/out folder by default. An example json array is shown below.

# Forward all images in test/ using tiny yolo and JSON output.
./flow --test test/ --model cfg/yolo-tiny.cfg --load bin/yolo-tiny.weights --json

JSON output:

[{"label":"person", "confidence": 0.56, "topleft": {"x": 184, "y": 101}, "bottomright": {"x": 274, "y": 382}},
{"label": "dog", "confidence": 0.32, "topleft": {"x": 71, "y": 263}, "bottomright": {"x": 193, "y": 353}},
{"label": "horse", "confidence": 0.76, "topleft": {"x": 412, "y": 109}, "bottomright": {"x": 592,"y": 337}}]

• label: self explanatory
• confidence: somewhere between 0 and 1 (how confident yolo is about that detection)
• topleft: pixel coordinate of top left corner of box.
• bottomright: pixel coordinate of bottom right corner of box.

For Training New Models

Training is simple as you only have to add option --train like below:

# Initialize yolo-3c from yolo-tiny, then train the net on 100% GPU:
./flow --model cfg/yolo-3c.cfg --load bin/yolo-tiny.weights --train --gpu 1.0

# Completely initialize yolo-3c and train it with ADAM optimizer
./flow --model cfg/yolo-3c.cfg --train --trainer adam

During training, the script will occasionally save intermediate results into Tensorflow checkpoints, stored in ckpt/. To resume to any checkpoint before performing training/testing, use --load [checkpoint_num] option, if checkpoint_num < 0, darkflow will load the most recent save by parsing ckpt/checkpoint.

# Resume the most recent checkpoint for training
./flow --train --model cfg/yolo-3c.cfg --load -1

# Test with checkpoint at step 1500
./flow --model cfg/yolo-3c.cfg --load 1500

# Fine tuning yolo-tiny from the original one
./flow --train --model cfg/yolo-tiny.cfg --load bin/yolo-tiny.weights