Object detection algorithm on jetson TK1 for real-time

Hi,

I am novice in deep learning and currently working on object detection on Jetson TK1 with real-time video from a webcam. The two frameworks or applications I tired are

Darknet Yolo: Got 5-6 frames per second with Tiny-yolo configuration and tiny-yolo weights
Single shot multibox detector (works on caffe): got 5 fps

Has anyone used these or any other framework for object detection (classification and localization of objects in an image) for real-time video on the Jetsons? If yes, then what was the maximum frame rate achieved?

Are there better frameworks for the same?

Any help would be greatly appreciated.

Thanks,
BS

Hi BharatS,
On TK1, could you try with max performance http://elinux.org/Jetson/Performance ?

We have samples of deep learning on tx1 but don’t have plan to make it to tk1. It would be good if you can try your case on tx1.

Hi DaneLLL,

Over clocking the system or using max performance did not improve the results. Yes, TX1 might help, but currently I am working only on TK1.

Hello, I saw your post and I am new to the jetson, I wanted to achieve the image/object detection using Yolo or SSD, can you please guide me through the process of achieving this? I know you said the max you achieved was 5-6 fps but this is something I’m really interested in and would like to at least get to that point :)

Hi,

Here are some relevant topics:

YOLO: (Yolo1 model can directly inference with TensorRT)
https://devtalk.nvidia.com/default/topic/990426/jetson-tx1/tensorrt-yolo-inference-error/

SSD: You need to enable their Caffe branch on TK1.
Please refer this: https://gist.github.com/jetsonhacks/acf63b993b44e1fb9528
Please remember to add TK1 GPU architecture into the Makefile.config
https://devtalk.nvidia.com/default/topic/1020850/jetson-tk1/tk1-python-caffe-problems/post/5196767/#5196767

Hello, i am trying to get object detection working on jetson using darknet yolo. Can you please guide me step by step on how you achieved the 5-6 fps configuration working?

Hi,

We don’t have experience with darknet. Instead, we use TensorRT to inference YOLO model.
Check this topic: https://devtalk.nvidia.com/default/topic/990426/jetson-tx1/tensorrt-yolo-inference-error/