I want to test deepstream-3d-action-recognition But I cann't find fall_floor

[fanzh]

please refer to my last comment. To narrow down this issue, if you only set the file name, can the “fall_floor” be detected? if yes, we need to check which new configuration modifications cause the issue.

[940655439]

can the “fall_floor” can not be detected.
the score is negative
The configuration is like this

[940655439]

I can find fall_floor .
There are two reasons for this.
One is the code.like below

only print last label

The second reason the model resnet18_3d_rgb_hmdb5_32.etlt
may have problem .
Now I use model resnet18_2d_rgb_hmdb5_32.etlt

[940655439]

The conclusion is that with this problem , the model resnet18_3d_rgb_hmdb5_32.etlt does not support fall_floor

[fanzh]

Thanks for sharing! I used deployable_v1.0. here is the md5sum value.
md5sum resnet18_3d_rgb_hmdb5_32.etlt
34122408d0ef94e87e3eb62485689614 resnet18_3d_rgb_hmdb5_32.etlt

[940655439]

I want to ask a questions about floor. How to get the coordinate of object , frameid, objectid and the picture of object ?
Did you help me deal with this need?

[fanzh]

In this sample, there is only one classification model. the object is the whole frame. you can get width, height and frame_num from NvDsFrameMeta of NvDsRoiMeta. please refer to function pgie_src_pad_buffer_probe of opt\nvidia\deepstream\deepstream\sources\apps\sample_apps\deepstream-3d-action-recognition\deepstream_3d_action_recognition.cpp.

[940655439]

I had saw this function many times, I Know what you said above.
In this function, you only get some message from batch_user_meta_list., not from frame_meta_list

I want to get more data.
1、 get the coordinate of fall_floor
2、 get the frameid of fall_floor

Because I need to process after inferring.

[fanzh]

from the last line " for (auto &roi_meta : preprocess_batchmeta->roi_vector)", you can get NvDsRoiMeta. then you can get width, height and frame_num from NvDsFrameMeta of NvDsRoiMeta.

[940655439]

1. Did you set the sink type 3 and save this video

[940655439]

I have another question,
Did you show in detail your testing?
I don’t know whether are difference between your and mine

[fanzh]

Thanks for the update. seems the original issue is fixed. could you help to open a new topic for the new problem? Thanks!