• State TI Thinks Resolved
  • Locked Locked
  • Replies 1 reply
  • Answers 1 answer
  • Subscribers 97 subscribers
  • Views 299 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

TDA2SX: TIDL performance issue using Jdetnet of GIT

Sung-IL
Mastermind 31520 points

Part Number: TDA2SX

Dear Champs,

My customer is trying to develop their own DL algorithm using Jacinto-Caffe, and they tried to convert pre-trained network found at below GIT directory.

https://git.ti.com/cgit/jacinto-ai-devkit/caffe-jacinto-models/tree/trained/object_detection/voc0712/JDetNet/ssd768x320_ds_PSP_dsFac_32_hdDS8_0/sparse

But, when they convert pre-trained network of JDETNET 768x320 found at GIT without any modification, they faced DSP hang and they can run it after modifying 'SYSN_THRESHOLD' value and found to run it at 2fps only.

I attached the log when DSP hang occurred as below. Could you please check below log and let me know what is the issue causing DSP hang?.

 debug_txt.txt

They can run it by changing 'SYNC_THRESHOLD' as below and they found 2fps at this time.

And also, they found the C66x DSP load is only 20% when they run JDETNET 768x320 included in visionSDK, and Please let me know how they can achieve similar performance using Jacinto-Caffe.

Thanks and Best Regards,

SI.