**** BEGIN LOGGING AT Fri Mar 05 02:59:57 2021 Mar 05 07:45:57 > lorforlinux: just to be clear and summarize the task I need to test the performance of yolo/tf lite models on rpi right? Mar 05 07:45:57 I completed the task Mar 05 07:46:17 I ran yolo v2 and yolo v3 models on rpi Mar 05 07:48:09 * I ran yolo v2(tiny) and yolo v3(tiny) models on rpi Mar 05 07:48:35 for the yolo v2 i was able to get an execution time that avg on 38s Mar 05 07:49:48 for the tiny yolo v3 models i was able to get an inference time that avg around 2.2s Mar 05 07:50:29 https://docs.google.com/document/d/1BFRIBIBc_wVopg0PhxrbWYh5kMI7BkuG8VlG5ifNkhc/edit?usp=sharing heres a detailed log of the entire process that I did Mar 05 07:50:42 @lor Mar 05 07:51:11 * lorforlinux,@vedant16 please do take a look thanks Mar 05 07:56:29 also in the beaglebone ideas list - https://elinux.org/BeagleBoard/GSoC/Ideas-2021 that provided details about the yolo models on X15/AI there were several work-arounds that were listed and considering the reduced 10 week (175 hr) time being provided for GSOC this year,I thinks its pretty reasonable that I work on porting yolo model into TIDL for the solution to this project Mar 05 08:03:20 After reading through TIDL api docs and user guides I found the general work through for this idea to use TIDL importer to first convert quantized model into .bin file and then use tidl apis to write a code that properly distributes the layer groups among the DSP and EVE blocks (similar to the segementation example for SSD) provided in TIDL docs Mar 05 08:03:41 > After reading through TIDL api docs and user guides I found the general work through for this idea to use TIDL importer to first convert quantized model into .bin file and then use tidl apis to write a code that properly distributes the layer groups among the DSP and EVE blocks (similar to the segementation example for SSD) provided in TIDL docs Mar 05 08:03:42 would like to the mentors' thoughts on this Mar 05 08:04:54 * > After reading through TIDL api docs and user guides I found the general work through for this idea to use TIDL importer to first convert quantized model into .bin file and then use tidl apis to write a code that properly distributes the layer groups among the DSP and EVE blocks (similar to the segementation example for SSD) provided in TIDL docs Mar 05 08:04:54 would like to know the mentors' thoughts on this Mar 05 08:06:52 also I was planning to setup the environment for the tidl as described here : http://software-dl.ti.com/jacinto7/esd/processor-sdk-rtos-jacinto7/06_01_01_12/exports/docs/tidl_j7_01_00_01_00/ti_dl/docs/user_guide_html/md_tidl_user_model_deployment.html but was confused with the SDK that I should download Mar 05 08:12:09 the one they have described is the jacinto 7 rtos and the one that I think would be required is the AM57x processor sdk .I was confused with which one to use.Would be great if someone could help me in this. Mar 05 08:13:55 > After reading through TIDL api docs and user guides I found the general work through for this idea to use TIDL importer to first convert quantized model into .bin file and then use tidl apis to write a code that properly distributes the layer groups among the DSP and EVE blocks (similar to the segementation example for SSD) provided in TIDL docs Mar 05 08:13:55 would like to know the mentors' thoughts on this Mar 05 08:13:55 I was planning to test this idea on smaller cnn models and in a way try to generalize this task Mar 05 08:14:10 Thanks ! **** ENDING LOGGING AT Sat Mar 06 02:59:57 2021