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.
Part Number: TDA2EVM5777
Hello,
My customer is trying to run TIDL test application on TDA2. They are using TIDL 01.01.03.00 in PROCESSOR SDK VISION 03.07.00.00.
(1)They want to import “ssd768x320_ds_PSP_dsFac_32_hdDS8_0” from below link.
https://git.ti.com/cgit/jacinto-ai-devkit/caffe-jacinto-models/tree/trained/object_detection/voc0712/JDetNet
We can find deploy.prottxt only in directory “test_quantize”. Can we use it for importing the network to TIDL?
(2)After importing the network by step (1), they compared the result with JDetNet structure on Web. It looks like there are some difference in network structure between them.
Example:
[Layer #] [Original network] [Imported network]
Layer No.8 Pooling Convolution
Layer No.10 Convolution Pooling
We also see the difference in parameter values. Please find attachment for details.
Could you please let us know the reason of the difference?
Do we need to do anything for that?
Best regards,
-Shibata
Hi Shibata,
(1) Yes, please use the deploy.prototxt from directory “test_quantize”.
(2) This is expected because both the model are different, the "JDetNet structure on Web" had only 4 classes, but this model in git has 21 classes.
For more details please refer to below e2e link.
https://e2e.ti.com/support/processors/f/791/t/689617
Thanks,
Praveen