Error using standalone image inference

hey there,

training worked fine, also making the inference models was no problem. I then build for standalone use, which gives me no errors.

But when I try using standalone inference on images as described in deploy/segmentation, I get the following C++ error:

Predicting image: ../train/tasks/segmentation/dataset/504-896/rgb/rgb-cropped/
terminate called after throwing an instance of 'std::out_of_range'
  what():  basic_string::substr: __pos (which is 140) > this->size() (which is 0)
Aborted (core dumped)

Since I dont yet have any idea of C++ whatsoever, I thought maybe someone has a chance to see the problem directly. In the following I'll add the whole console output for my infer_img execution, aswell as the config of the model. I would appreciate any clue. Thx for looking!

My example is called with tensor rt, but I get the same error with pytorch aswell. I first thought it was about image dimensions, remade the inference models with new dimensions etc., but did not get there yet.

The full console output:

developer@olli:/home/olli/Code/XxxXxxx/bonnetal/deploy$ ./devel/lib/bonnetal_segmentation_standalone/infer_img -p ../train/tasks/segmentation/mystuff/deployed/best_darknet_leaf -i ../train/tasks/segmentation/dataset/504-896/rgb/rgb-cropped/ -b tensorrt
================================================================================
image: ../train/tasks/segmentation/dataset/504-896/rgb/rgb-cropped/
path: ../train/tasks/segmentation/mystuff/deployed/best_darknet_leaf/
backend: tensorrt
verbose: 0
================================================================================
Setting verbosity to: true
Trying to open model
Trying to deserialize previously stored: ../train/tasks/segmentation/mystuff/deployed/best_darknet_leaf//model.trt
Successfully found TensorRT engine file ../train/tasks/segmentation/mystuff/deployed/best_darknet_leaf//model.trt
Successfully created inference runtime
No DLA selected.
Successfully allocated 122542320 for model.
Successfully read 122542320 to modelmem.
INFO: Glob Size is 122257600 bytes.
INFO: Added linear block of size 173408256
INFO: Added linear block of size 173408256
INFO: Added linear block of size 28901376
INFO: Added linear block of size 28901376
INFO: Added linear block of size 14450688
INFO: Added linear block of size 7225344
INFO: Deserialize required 1790698 microseconds.
Created engine!
Successfully deserialized Engine from trt file
Binding: 0, type: 0
[Dim 3][Dim 504][Dim 896]
Binding: 1, type: 3
[Dim 1][Dim 504][Dim 896]
Successfully create binding buffer
================================================================================
Predicting image: ../train/tasks/segmentation/dataset/504-896/rgb/rgb-cropped/
terminate called after throwing an instance of 'std::out_of_range'
  what():  basic_string::substr: __pos (which is 140) > this->size() (which is 0)
Aborted (core dumped)

And the config

backbone:
  OS: 8
  bn_d: 0.01
  dropout: 0.01
  extra:
    darknet: darknet53
  name: darknet
  train: true
dataset:
  color_map:
    0:
    - 0
    - 0
    - 0
    1:
    - 255
    - 0
    - 0
    2:
    - 0
    - 0
    - 255
  img_means:
  - 0.4148446751432425
  - 0.5053385609354691
  - 0.45907718096515426
  img_prop:
    depth: 3
    height: 504
    width: 896
  img_stds:
  - 0.14722864867881855
  - 0.16334236156069035
  - 0.17758600209156641
  labels:
    0: ground
    1: carrot
    2: weed
  labels_w:
    0: 1.0
    1: 1.0
    2: 1.0
  location: dataset/leaf_moreweed/
  name: leaf_moreweed
  workers: 12
decoder:
  bn_d: 0.01
  dropout: 0.01
  extra:
    aspp_channels: 256
    last_channels: 32
    skip_os:
    - 4
    - 2
  name: aspp_residual
  train: true
head:
  dropout: 0.01
  name: segmentation
train:
  avg_N: 3
  batch_size: 2
  crop_prop:
    height: 480
    width: 480
  down_epochs: 100
  final_decay: 0.995
  loss: xentropy
  max_epochs: 100
  max_lr: 0.0001
  max_momentum: 0.95
  min_lr: 1.0e-05
  min_momentum: 0.9
  report_batch: 1
  report_epoch: 1
  save_imgs: false
  save_summary: true
  up_epochs: 0.5
  w_decay: 0.0001

best regards, olli

PRBonn / bonnetal

Error using standalone image inference #28