Question about Traffic Light Detection Using the TensorFlow* Object Detection API

[TaejinKim]

Hi, I'm trying to build traffic light detection model with tensorflow object detection api.

I've found this link(https://www.intel.com/content/www/us/en/developer/articles/technical/traffic-light-detection-using-the-tensorflow-object-detection-api.html)

and tried to follow the steps.

I couldn't find any imagenet dataset(http://imagenet.stanford.edu/synset?wnid=n06874185)

Is there any updates on this link(https://www.intel.com/content/www/us/en/developer/articles/technical/traffic-light-detection-using-the-tensorflow-object-detection-api.html)?

Regards,

Jason.

[Megat_Intel · Accepted solution]

Hi Jason,

Thank you for reaching out to us.

For the ImageNet dataset, I found that Large Scale Visual Recognition Challenge 2017 (ILSVRC2017) contains the synsets for (n06874185: traffic light) from here. You can download the ILSVRC2017 dataset here. Please note that you need to download the entire DET dataset which contains all 200 synsets for object detection, including traffic light.

Also, the link that you shared was last updated in April 2018 and was a pre-OpenVINO era tutorial. There are no plans for updates right now, and we encourage you to use Intel® Distribution of OpenVINO™ Toolkit for optimizing and deploying AI inference.

We also suggest you explore Open Model Zoo, which contains over 200 neural network models, both public and from Intel, and covers wide application areas, including object detection, classification, image segmentation, etc. The Intel models have already been converted to work with OpenVINO™ toolkit, while public models can easily be converted using Model Optimizer.

You can also use Open Model Zoo Demos, which provide robust application templates to help you implement specific deep learning scenarios. 

Regards,

Megat

[TaejinKim]

Hi Megat,

Thank you for your answer with explanation.

I know there are lots of pre-trained model provided by OpenVINO, but I couldn't find model that detects traffic light. And also there are lots of object detection models, so is it better to check classes from these models?

If you don't mind, could you recommend me any model?

Regards,

Jason.

[TaejinKim]

Hi Megat,

Don't mind on my previous question about recommend models.

I've found yolov3 to detect traffic light!

Now I have to work on recognize color of it.

Regards,

Jason.

[Megat_Intel]

Hi Jason,

Nice to know that. On my side, I've proactively searched the Open Model Zoo for relevant models that were trained with datasets containing traffic light classes. I then validated those models by running Object Detection C++ Demo / Object Detection Python* Demo, to verify whether they are able to detect traffic lights.

The models in Open Model Zoo that can detect traffic lights are as follows: 

1. OpenVINO™ Toolkit Intel’s Pre-Trained Model:

• faster-rcnn-resnet101-coco-sparse-60-0001

2. OpenVINO™ Toolkit Public Pre-Trained Models:

• ssdlite_mobilenet_v2
• ssd_mobilenet_v1_coco
• ctdet_coco_dlav0_512
• efficientdet-d0-tf
• efficientdet-d1-tf
• faster_rcnn_inception_resnet_v2_atrous_coco
• faster_rcnn_resnet50_coco
• rfcn-resnet101-coco-tf
• pelee-coco
• ssd-resnet34-1200-onnx
• yolo-v2-tiny-tf
• yolo-v3-tiny-tf
• yolo-v2-tf
• yolo-v3-tf
• yolof
• yolo-v3-tiny-onnx
• yolo-v3-onnx
• yolox-tiny

Regards,

Megat

[Megat_Intel]

Hi Jason,

This thread will no longer be monitored since we have provided a suggestion. If you need any additional information from Intel, please submit a new question.

Regards,

Megat