Using Tappas on a custom segformer hef model

General
, ,
1

Hello,
I have a segformer b0 which i manage to compile. I ahve its hef version and running $hailortcli run my_segformer_b0.hef works fine.
No I wish to see its performance. I installed the latest tappas and introduce to it the semantic segmentation from previous versions of the tappas (because the latest tappas doesnt include segmentation) which also works fine. Now, when I am trying to use my_segformer_b0.hef in the tappas i am getting an error. here is the output while trying to run my model from the sematic_segmentation.sh inside the tappas…

(mytappas) nuchailo1@nuchailo1:~/Documents/tappas30_manual/tappas_v3.30.0/apps/h8/gstreamer/general/segmentation$ ./semantic_segmentation.sh
Running /home/nuchailo1/Documents/tappas30_manual/tappas_v3.30.0/apps/h8/gstreamer/general/segmentation/resources/b0_best_mIoU_iter_212000_opset15_sim_480_848.hef
gst-launch-1.0 filesrc location=/home/nuchailo1/Documents/tappas30_manual/tappas_v3.30.0/apps/h8/gstreamer/general/segmentation/resources/captured_video.mp4 ! decodebin ! videoscale ! video/x-raw, pixel-aspect-ratio=1/1 ! videoconvert ! queue leaky=no max-size-buffers=30 max-size-bytes=0 max-size-time=0 ! hailonet hef-path=/home/nuchailo1/Documents/tappas30_manual/tappas_v3.30.0/apps/h8/gstreamer/general/segmentation/resources/b0_best_mIoU_iter_212000_opset15_sim_480_848.hef ! queue leaky=no max-size-buffers=30 max-size-bytes=0 max-size-time=0 ! hailofilter so-path=/home/nuchailo1/Documents/tappas30_manual/tappas_v3.30.0/apps/h8/gstreamer/libs/post_processes//libsemantic_segmentation.so qos=false ! queue leaky=no max-size-buffers=30 max-size-bytes=0 max-size-time=0 ! hailooverlay qos=false ! queue leaky=no max-size-buffers=30 max-size-bytes=0 max-size-time=0 ! videoconvert ! fpsdisplaysink video-sink=xvimagesink name=hailo_display sync=false text-overlay=false
Setting pipeline to PAUSED …
Pipeline is PREROLLING …
Redistribute latency…
Redistribute latency…
HailoNet Error: gst_pad_push failed with status = -4

p.s - my model accept images with one channel. grayscale and i am giving him a video of images having one channel.
Any Ideas?

David

2

Hi @dudibs,
The TAPPAS component is used for demos and examples, and therefore is not generic. You can only run the TAPPAS examples with the corresponding HEF file for the designated task. Unless your model is similar to the model of the example you are trying to run, you won’t be able to use the existing examples as it matches the model to it’s postprocessing, for example.

You’ll need to build your own pipeline, meaning create the relevant postprocess .so to be able to run it with TAPPAS.

Regards,