sample_embedding_folder2/0807726.txt

Ticket Name: TDA2EVM5777: Cannot see any output of semantic segmentation, trained based on the Caffe-Jacinto's Jsegnet

Query Text:
Part Number: TDA2EVM5777 Other Parts Discussed in Thread: TDA2 Hi, After successful training and testing on the PC till the sparsification stage, I used the import tool to to generate the net and prm files of jsegnet. Converted the video as well which is also based on my previous experience of testing the demo version. However, when I run these files on the tda2, all I can see is the static output frame of my video which isn't moving at all. I didn't see any error while importation and execution, so I assume that there is no problem in the network. I used the semseg usecase with default settings to run my trained model and input video, however, my model is trained on two classes rather than four as I have only two objects. Things I noticed while execution: Cannot allocate the object for sync1,2,3 and algorithm 0,1,2,3,...7. Other than this, I cannot find any possible error. Another thing is that when I pause and resume my video it runs for that particular interval without any segmentation result otherwise it is stuck. Here is the log file.2330.log.txt. Kindly respond me as soon as possible. Thank you. With best regards, H.M. Owais

Responses:
Hi, Please refer to below two e2e threads for training and running custom data set using semantic segmentation use case e2e.ti.com/.../660676 e2e.ti.com/.../794965 Thanks, Praveen

Hi Praveen, Thank you for your reply, however, I have already seen these threads and couldn't find any relevant solution after searching a lot on e2e threads. I am dealing with different problem. I can try to explain it again. 1. I have done successful training with my own dataset (not using cityscape). My dataset has only two objects which need pixel-wise classification. 2. I used caffe-jacinto to train and tested on the PC successfully. My test dataset shows output successfully on the PC. 3. I converted the sparsed model using the import tool and generated BIN files for parameter and network. The import tool showed successful importation and generation of files. 4. I have also converted the test video as per the given instructions. My test video works well with other models. 5. I am using the same usecase, semSeg given in the demo codes with all default settings to display the results. This is what I have done so far. Now here are my questions, 1. Why I can't see the output of my two objects? it just shows me the test video frame, nothing else. 2. Where I can find the network output in the tidl_algplugin.c? can you precisely tell me the function to display it?\ 3. I have also attached the log file in the above question, can you have a look at it and tell me where the problem is ? 4.. If there is any helpful document which contains all the information about tidl_algplugin.c. Kindly share it with me as well. Thank you. With best regards, H.M. Owais

Hi, From the log it seems sync link is dropping frame. Can you increase SYNC_THRESHOLD to 3000 is your semseg usecase. Regards, Anuj

Hi, Its been long since any update on this. I hope the issue is resolved, if not then please reply below or create anew thread for different issue. Regards, Anuj

Hi Anuj, Sorry for being late in replying the answer. The issue was with the input inData file. I resolved it back then. Thank you for your response. With best regards, H.M. Owais

Hi, You can mark any answer as it resolved your thread and close it. Regards, Anuj