Seems you are using Jetson. Please try deepstream_image_meta_test.c (20.3 KB)
And please notice there are some issue with deepstream-image-meta-test sample now with Jetson. Deepstream-transfer-learning-app saving wrong image file name - Intelligent Video Analytics / DeepStream SDK - NVIDIA Developer Forums
If we can reproduce the problem. We can debug the bug. The information mean nothing to you since the implementation is not open source.
I find that the deepstream-image-meta-test.c file you sent me this time is no different from that last time.I’ve tested this file this afternoon.And I didn’t encounter the problem of green images when I used deepstream-image-meta-test sample now with Jetson.In addition, did your previous test result use the Jetson NX platform？
Sorry for sending the wrong file. Please try this file. deepstream_image_meta_test.c (20.3 KB)
I used dGPU, the code is just the same.
how about config source file
Hello.I used the deepstream_image_meta_test.c file that you sent yesterday to test.But saved images are still wrong.Can you test your sample code on the platform of Jetson NX and See if it can be reproduced.This problem is quite urgent. Please deal with it as soon as possible，thank you.
are you try this
Hello,Is there any conclusion to this question？
I can confirm the bug in a custom written gst plugin. As long as the batch size is 1 or the frame references the first surface in the batch, the image content is correct. If not, the cropped image content from the first surface (batch_id 0) is still returned.
So, can anybody with source code access from NVIDIA check the implementation of nvds_obj_enc_process? It should honor the batch_id from frame_meta, but it always seems to take the first surface in the batch.
Thanks
This bug fix will be integrated to the next DS release.
