sample-apps-for-robotics-platforms icon indicating copy to clipboard operation
sample-apps-for-robotics-platforms copied to clipboard

Could not run custom TFLite models on RB5 with GStreamer framework.

Open veeramallirajesh opened this issue 4 years ago • 0 comments

I was following the GStreamer TFLite object detection and segmentation examples on the repo. I could run them on RB5. But, when I try to run my own TFLite segmentation model on it the inference is not working. I tried to debug the "qtimletflite" element and what I found from it is the error "MLE init failed." When I include "org.codeaurora.mle.tflite" in the config file for qtimletflite element I get the following debug message. I would like to know if there are any limitations on how a TFLite model should be developed and compiled.

(blurnet_segmentation:2495): GStreamer-CRITICAL **: 09:08:06.041: gst_structure_new_empty: assertion 'gst_structure_validate_name (name)' failed 0:00:02.511495398 2495 0x55bc1f89e0 WARN qtimletflite mle_tflite.cc:308:gst_mle_tflite_parse_config: Failed to deserialize the config! 0:00:02.511622530 2495 0x55bc1f89e0 DEBUG qtimletflite mle_tflite.cc:414:gst_mle_create_engine: Parsing configuration failed. 0:00:02.511758933 2495 0x55bc1f89e0 DEBUG qtimletflite mle_tflite.cc:367:gst_mle_print_config: ==== Configuration Begin ==== 0:00:02.511897054 2495 0x55bc1f89e0 DEBUG qtimletflite mle_tflite.cc:368:gst_mle_print_config: Model /data/misc/camera/blurnet/blurnet_8.tflite 0:00:02.512022312 2495 0x55bc1f89e0 DEBUG qtimletflite mle_tflite.cc:369:gst_mle_print_config: Labels 0:00:02.512172099 2495 0x55bc1f89e0 DEBUG qtimletflite mle_tflite.cc:371:gst_mle_print_config: Pre-processing 1 0:00:02.512320950 2495 0x55bc1f89e0 DEBUG qtimletflite mle_tflite.cc:374:gst_mle_print_config: Mean(B,G,R): 0.000000, 0.000000, 0.000000 0:00:02.512455946 2495 0x55bc1f89e0 DEBUG qtimletflite mle_tflite.cc:377:gst_mle_print_config: Sigma(B,G,R): 0.000000, 0.000000, 0.000000 0:00:02.512584745 2495 0x55bc1f89e0 DEBUG qtimletflite mle_tflite.cc:379:gst_mle_print_config: Confidence threshold 0.500000 0:00:02.512706825 2495 0x55bc1f89e0 DEBUG qtimletflite mle_tflite.cc:380:gst_mle_print_config: Input format 0 0:00:02.512829686 2495 0x55bc1f89e0 DEBUG qtimletflite mle_tflite.cc:381:gst_mle_print_config: Delegate default 0:00:02.513061451 2495 0x55bc1f89e0 DEBUG qtimletflite mle_tflite.cc:383:gst_mle_print_config: Number of threads 2 0:00:02.513146345 2495 0x55bc1f89e0 DEBUG qtimletflite mle_tflite.cc:385:gst_mle_print_config: Post-processing segmentation 0:00:02.513220562 2495 0x55bc1f89e0 DEBUG qtimletflite mle_tflite.cc:386:gst_mle_print_config: ==== Configuration End ==== ERROR: Didn't find op for builtin opcode 'CONV_2D' version '5'

ERROR: Registration failed.

Thank you.

Regards Rajesh Veeramalli

veeramallirajesh avatar Jan 28 '21 09:01 veeramallirajesh