Yes it looks some improvements are there.
My second pgie’s input batch is supposed to be dynamic with max batch size is 10.
When updated according to your suggestions, the following are my observations.
Test with deepstream-app.
The outputs are
KLT Tracker Init
mInputDims from enqueue 88 10 48
mOutputDims from enqueue 10 20
before print
26 20 29 19 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 20 29 19 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 20 26 20 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 20 26 20 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 20 26 20 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 11 11 11 11 11 11 11 11 11 11 11 11 11 -1 -1 -1 -1 -1 -1 26 11 11 11 11 11 11 11 11 11 11 11 11 14 -1 -1 -1 -1 -1 -1 26 11 11 11 11 11 11 11 11 11 11 11 11 14 -1 -1 -1 -1 -1 -1 26 11 11 11 11 11 11 11 11 11 11 11 3 14 -1 -1 -1 -1 -1 -1 26 11 11 11 11 11 11 11 11 11 11 -1 -1 -1 -1 -1 -1 -1 -1 -1
dims 20
decoded as 020202020
dims 20
decoded as 0
dims 20
decoded as 020202020
dims 20
decoded as 0
dims 20
decoded as 020202020
dims 20
decoded as 0
dims 20
decoded as 020202020
dims 20
decoded as 0
dims 20
decoded as 020202020
dims 20
decoded as 0
mInputDims from enqueue 88 10 48
mOutputDims from enqueue 10 20
before print
26 11 11 11 11 11 11 11 11 11 11 11 -1 -1 -1 -1 -1 -1 -1 -1 26 11 11 11 11 11 11 11 11 11 11 14 -1 -1 -1 -1 -1 -1 -1 -1 26 11 11 11 11 11 11 11 11 11 11 11 -1 -1 -1 -1 -1 -1 -1 -1 26 11 11 11 11 11 11 11 11 11 11 11 -1 -1 -1 -1 -1 -1 -1 -1 26 11 11 11 11 11 11 11 11 11 11 11 -1 -1 -1 -1 -1 -1 -1 -1 26 11 11 11 11 11 8 11 11 11 11 11 11 14 -1 -1 -1 -1 -1 -1 26 11 14 11 14 11 11 11 11 11 11 11 14 -1 -1 -1 -1 -1 -1 -1 26 19 26 17 5 5 5 9 5 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 19 17 9 17 5 9 9 16 5 9 9 -1 -1 -1 -1 -1 -1 -1 -1 26 19 26 6 9 6 9 26 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1
dims 20
decoded as 02020202020202020202
dims 20
decoded as 02020
dims 20
decoded as 02020202020202020202
dims 20
decoded as 02020
dims 20
decoded as 02020202020202020202
dims 20
decoded as 02020
dims 20
decoded as 02020202020202020202
dims 20
decoded as 02020
dims 20
decoded as 02020202020202020202
dims 20
decoded as 02020
mInputDims from enqueue 88 10 48
mOutputDims from enqueue 10 20
before print
26 17 5 5 6 9 5 6 9 17 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 19 26 9 5 5 9 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 17 26 9 9 5 5 26 17 9 6 8 9 -1 -1 -1 -1 -1 -1 -1 26 17 26 9 9 5 5 26 17 9 6 8 9 -1 -1 -1 -1 -1 -1 -1 26 19 17 8 6 6 17 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 19 17 17 5 9 5 9 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 19 17 9 9 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 19 26 6 9 6 9 26 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 17 26 6 5 9 17 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 17 26 9 9 5 5 26 17 9 6 8 9 -1 -1 -1 -1 -1 -1 -1
dims 20
decoded as 02020202020202020202
dims 20
decoded as 0
dims 20
decoded as 020202020202020
dims 20
decoded as 0
dims 20
decoded as 02020202020202020202
dims 20
decoded as 0202020
dims 20
decoded as 02020202020202020202
dims 20
decoded as 0202020
dims 20
decoded as 020202020202020
dims 20
decoded as 0
mInputDims from enqueue 88 2 48
mOutputDims from enqueue 2 20
before print
26 19 26 17 9 5 6 9 5 9 9 16 -1 -1 -1 -1 -1 -1 -1 -1 26 19 26 17 9 5 6 9 5 9 9 16 -1 -1 -1 -1 -1 -1 -1 -1
dims 20
decoded as 02020202020202020202
dims 20
decoded as 02020
q
Quitting
Observations
With earlier setting batch-size=1, the program freezed and I need to reset AGX Xavier.
After setting batch-size=10, now is program can run till Q-key is pressed. So that is improvement.
Primary pgie’s detection output is not 10.
Sometimes 3 sometimes 4,5,6 etc. But never reached to 10.
(1)Why batch size to custom plugin is 10?
Print from custom plugin
mInputDims from enqueue 88 10 48
mOutputDims from enqueue 10 20
26 20 29 19 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 20 29 19 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 20 26 20 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 20 26 20 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 20 26 20 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 11 11 11 11 11 11 11 11 11 11 11 11 11 -1 -1 -1 -1 -1 -1 26 11 11 11 11 11 11 11 11 11 11 11 11 14 -1 -1 -1 -1 -1 -1 26 11 11 11 11 11 11 11 11 11 11 11 11 14 -1 -1 -1 -1 -1 -1 26 11 11 11 11 11 11 11 11 11 11 11 3 14 -1 -1 -1 -1 -1 -1 26 11 11 11 11 11 11 11 11 11 11 -1 -1 -1 -1 -1 -1 -1 -1 -1
It looks repetition of inputs to sgie.
When tested with deepstream-test2 with pgie’s batch-size=10, batch size is correct. So dynamic batch size changed according to number of detection at pgie. But still have error at custom plugin CDCGreedyDecoder’s outputs. 1st is correct, 2nd and 3rd are wrong.
mInputDims from enqueue 88 3 48
mOutputDims from enqueue 3 20
before print
26 20 26 26 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 26 11 26 11 11 11 11 11 11 11 11 11 14 -1 -1 -1 -1 -1 -1 -1 26 19 26 17 9 8 17 5 9 5 4 17 9 26 -1 -1 -1 -1 -1 -1
dims 20
decoded as 020202020
dims 20
decoded as 0
dims 20
decoded as 02020202020202020202
Input at custom parser also wrong, not same as CDCGreedyDecoder’s output.