Skip to content
Snippets Groups Projects
Select Git revision
  • master default protected
  • new_code
2 results
Compare
Name Last commit Last update
.vscode
cfg
data
models
onnx-tensorrt
utils
weights
.gitignore
Agenso_api.py
Agenso_thread.py
GPS_receiver.py
Install_Guide_Adlink_Neon.md
LICENSE
README.md
camera.py
data_types.py
detect.py
global_vars.py
hubconf.py
inference.py
requirements.txt
test.py
train.py
tutorial.ipynb
usb_Filestorage.py
README.md

Inference source-code for the Adlink Neon smart-camera

Installation

See Install_Guide_Adlink_Neon.md

Getting started

First copy-paste the ONNX weight-files into the folder that belongs to the specific use-case:

  • weights/apple_scab
  • weights/downy_mildew
  • weights/alternaria

Copy-paste both the _efficientnet_bs1.onnx and _efficientnet_bs6.onnx so that we can choose whether to analyse on the single image (bs1) or on the single image including five cropped (zoomed) patches of the image (bs6).

Image inference with the smart-camera

  1. Open a new terminal.
  2. Go to the root-folder (cd inference)
  3. Run the python file inference.py, using the appropriate arguments (see below):

Argument Description
--dl-mode Specify the deep-learning mode. Choose between detection or classification
--img-patches Give this command when you want to analyse the image including five cropped (zoomed) patches of the image
--use-case What use-case are you working on? Choose apple_scab or downy_mildew or alternaria
--img-size The input image-size for the deep-learning algorithm.
--max-img-width Optional: when img-patches is activated, specify the desired maximum width to transform the image to
--max-img-height Optional: when img-patches is activated, specify the desired maximum height to transform the image to
--source Optional: input folder that contains the images to read
--weightsfolder Folder with the weight-files. Default is weights
--device Specify the computing device. Choose between CUDA or cpu
--conf-thres Confidence-threshold for the detection algorithm
--nms-thres IOU threshold for non-maximum suppression for the detection algorithm
--output Neural net output folder. Default is results/adlink_acquistion_session1
--view-img Give this command when you want to display the images and the detections/classes
--logfile-folder Folder save the logfiles in. Default is data/temp
--img-folder Folder to save image in. Default is data/temp/img
--data-folder Folder to save the GPS data in. Default is data/temp/GPSdata
--tosend-folder Folder to save unsend jsons in. Default is data/temp/tosend
--badreq-folder Folder to save jsons that give errors in. Default is data/temp/badreq
--debug-mode Give this command when you want to read images from source instead of camera
--disable-data Give this command when you want to disable the reading of the txt detections files
--nn-only Give this command when you want to only run neural network with images from the source for testing purposes
--tensorrt Give this command when you want to activate TensorRT (to speed-up the algorithm)

Example:
python inference.py --dl-mode detection --use-case apple_scab --img-size 640 --conf-thres 0.4 --nms-thres 0.3 --view-img --tensorrt


Image inference with images loaded from disk (for testing purposes)

  1. Open a new terminal.
  2. Go to the annotation folder (cd inference)
  3. Run the python file inference.py, using the appropriate arguments (see above)

Example:
python inference.py --dl-mode classification --img-patches --use-case apple_scab --img-size 545 --source data/images --view-img --debug-mode --disable-data --nn-only --tensorrt


Contribution

This source-code was developed for the H2020 OPTIMA project.

Please contact us when you have some questions: