sample_embedding_folder2/0762118.txt

Ticket Name: TDA2: Deep learning algorithms run very slowly on DSP(TDA2xx)

Query Text:
Part Number: TDA2 Hi, Deep learning algorithms run very slowly on DSP(TDA2xx) image size is 512x256. ### CPU [ DSP1], LinkID [ 49], [IPU1-0] 831.681815 s: [IPU1-0] 831.681845 s: [ ALG_TIDL ] Link Statistics, [IPU1-0] 831.681937 s: ****************************** [IPU1-0] 831.681998 s: [IPU1-0] 831.682028 s: Elapsed time = 633812 msec [IPU1-0] 831.682120 s: [IPU1-0] 831.682272 s: New data Recv = 0.2 fps [IPU1-0] 831.682364 s: [IPU1-0] 831.682394 s: Input Statistics, [IPU1-0] 831.682455 s: [IPU1-0] 831.682516 s: CH | In Recv | In Drop | In User Drop | In Process [IPU1-0] 831.682577 s: | FPS | FPS | FPS | FPS [IPU1-0] 831.682669 s: -------------------------------------------------- [IPU1-0] 831.682760 s: 0 | 0. 1 0. 0 0. 0 0. 1 Computational complexity: thanks!

Responses:
Hi, can you check this post by Kumar and update "deploy.prototxt " parameters as described and see if it improves the speed: e2e.ti.com/.../689617 Regards, Yordan

Hi, deploy_512x256.rarAbout eep_top_k: 20 confidence_threshold: 0.15 1. I don't have these two variables in my file (deploy.prototxt) . 2. I want to run four networks on tda2x at the same time. 1camera (video input)---------->first network on DSP_1 2camera(video input)---------->the second network on DSP_2 3camera(video input)----------->the third network on EVE_1 4camera(video input)---------->the fourth network on EVE_2 EVE_3 EVE_4 The network is running slowly on the DSP. Is there any way to improve it? Thanks!

Refer The data sheet for performance comparision between EVE and DSP. We dont see much scope the improve the performnce of DSP