How to fix green image save by deepstream-transfer-learning-app?

[PhongNT]

To fix green image I found the solution at I need to send the image with bbox and other metadata to the custom server.

@amycao

There one temporary solution. adding nvvideoconvert component before a component where jpeg encode takes place. Set colorimetry to 601 in the caps of nvvideoconvert

i checked path in this link, this path custom GstElement pipeline to add nvvidconv1 and cap_filter. But this solution work deepstream-image-meta-test with GstElement pipeline
So in deepstream-transfer-learning-app i can’t see GstElement pipeline

How to apply this custom to deepstream-transfer-learning-app ?
Help me!

• Hardware Platform Jetson NX
• DeepStream Version 5.1
• JetPack Version 4.4
• How to reproduce the issue ?

[Fiona.Chen]

@PhongNT

So in deepstream-transfer-learning-app i can’t see GstElement pipeline

Why do you say so. There is “create_pipeline()” function in deepstream-transfer-learning-app source code, this is the pipeline.

Please read the source code carefully.
Please use the following two files to replace /opt/nvidia/deepstream/deepstream/sources/apps/apps_common/src/deepstream_primary_gie_bin.c and /opt/nvidia/deepstream/deepstream/sources/apps/apps_common/includes/deepstream_primary_gie.h and then rebuild deepstream-transfer-learning-app
deepstream_primary_gie.h (2.0 KB)
deepstream_primary_gie_bin.c (6.5 KB)

[PhongNT]

hello, it sill green image
are you try success with your patch in deepstream-transfer-learning-app ?

[Fiona.Chen]

Yes. It run successfully with deepstream-transfer-learning-app in Jetson.

[PhongNT]

are you use default config file in deepstream-transfer-learning-app and not change source video ?

[Fiona.Chen]

Surely yes. Only modify the two files I posted in this topic.

[PhongNT]

thank you so much