Some problems about deepstream_test2 demo

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Hello, I want to test the cascade model on deepstream. I reference the deepstream_test2, but encountered some problems:

I find the secondary model only detect part of the content from the primary model. Such as “Car 786”, but “Car737、Car792 …” have no contens. You can see follow picture:

Screenshot from 2020-09-27 14-42-52
Screenshot from 2020-09-27 14-42-521920×1080 576 KB

Then I print out the information of secondary model and part of the “car Bboxes” have contents.

Screenshot from 2020-09-27 15-04-36
Screenshot from 2020-09-27 15-04-36731×292 91.4 KB

Now, I want all the “car Bboxes” send to the secondary model and get each of them output categories. How can I do ?

3

please disable the tracker and try agian.

Thanks!

4

@mchi Hello, I have disabled the tracker. None “car Bboxes” have classification results.

Screenshot from 2020-09-29 10-15-40
Screenshot from 2020-09-29 10-15-401920×1080 579 KB

5

HI,
Reproed your issue, and we are checking internally. thanks for your patience.

6

@Amycao @kayccc @mchi Hello， Have results of your discussion?

7

Sorry for late reply, we are on holiday, and we are checking, thanks for your patience.

8

Hi @383109759,
could you try lower the “classifier-threshold” of sgies?

Thanks!

9

@mchi I had tried lower the “classifier-threshold” of sgies, such as 0.01. It seems have no effect. Dose you lower the threshold have some effect？

10

Please try set below properties in secondary gies config,
input-object-min-width=0
input-object-min-height=0
it will classify most of the objects.

11

@Amycao Hi, I have tried set below properties in secondary gies config:
input-object-min-width=0
input-object-min-height=0
But it only has one or two “car BBox” was classified. It is the same as before. Did I not set it correctly?

When I disable the tracker, None of the “car BBox” are classified, It seems to related to the tracker.

12

Please see log, most of the objects classified, before the config change, a few can be classified.
log.txt (707.9 KB)
btw, I am using deepstream-infer-tensor-meta-test, but should be similar.

1 Like
13

@Amycao I used “deepstream-infer-tensor-meta-test” and set “input-object-min-width=0,input-object-min-height=0”, most of objects classified.
Thank you very much!