Object Detection with YOLOX (PyTorch) - training with python in Colab, inference with C++
Budget: €30 – €250 EUR
SAMPLE DATA
===========
Sample images are in the attached zip d02.zip under
CreatedImages
The annotations are given in the text files
df_out_test.txt
df_out_train.txt
Suppose data is in Google Drive under
'/content/gdrive/MyDrive/SampleObjectDetection02/od01/'
So start with an Colab notebook like:
from google.colab import drive
drive.mount("/content/gdrive", force_remount=True)
dir = '/content/gdrive/MyDrive/SampleObjectDetection02/od01/'
%cd {dir}
TASKS
=====
I) Training in Colab
--------------------
Remark: This all has to be done in Colab. (For development you nust ahve a Colab environment on your own!)
1) Prepare data for use in training using a pretrained YOLOX model
2) Train model and show how to set the following training features:
- selection of pretrained model (YOLOX-s, YOLOX-x,...)
- settings for data augmentation, at least :
* rotation with limits for angle
* flipping vertical and horizontal
* brightness and/or contrast with limits for brightness/contrat
* zoom with limits for zoom factor
- how to freeze layers
- use of tensorboard
3) Evaluate model, calculate measures for goodness of model abritray test data
4) Inference: Use model with a specific image, show results (rectangles around objects found), save results as tab-delimited text file
-> Send notebook which I can run in Colab
II) Inference with C++
----------------------
Remark: It is fine to use OpenCV (C++) via vcpkg for this part.
1) Setup a concole C++ Visual Studio 2021 project (or Embarcadero C++ Builder 11 project)
2) Use trained model from above
3) Calculate specific image, show results (rectangles around objects found), save results as tab-delimited text file
-> Send source code for C++ project
Both tasks have to be done sucessfully on my machine before payment is released. Of course, I have tp be able to perform the training in my Colab and build and run the C++ project for inference.
===========
Sample images are in the attached zip d02.zip under
CreatedImages
The annotations are given in the text files
df_out_test.txt
df_out_train.txt
Suppose data is in Google Drive under
'/content/gdrive/MyDrive/SampleObjectDetection02/od01/'
So start with an Colab notebook like:
from google.colab import drive
drive.mount("/content/gdrive", force_remount=True)
dir = '/content/gdrive/MyDrive/SampleObjectDetection02/od01/'
%cd {dir}
TASKS
=====
I) Training in Colab
--------------------
Remark: This all has to be done in Colab. (For development you nust ahve a Colab environment on your own!)
1) Prepare data for use in training using a pretrained YOLOX model
2) Train model and show how to set the following training features:
- selection of pretrained model (YOLOX-s, YOLOX-x,...)
- settings for data augmentation, at least :
* rotation with limits for angle
* flipping vertical and horizontal
* brightness and/or contrast with limits for brightness/contrat
* zoom with limits for zoom factor
- how to freeze layers
- use of tensorboard
3) Evaluate model, calculate measures for goodness of model abritray test data
4) Inference: Use model with a specific image, show results (rectangles around objects found), save results as tab-delimited text file
-> Send notebook which I can run in Colab
II) Inference with C++
----------------------
Remark: It is fine to use OpenCV (C++) via vcpkg for this part.
1) Setup a concole C++ Visual Studio 2021 project (or Embarcadero C++ Builder 11 project)
2) Use trained model from above
3) Calculate specific image, show results (rectangles around objects found), save results as tab-delimited text file
-> Send source code for C++ project
Both tasks have to be done sucessfully on my machine before payment is released. Of course, I have tp be able to perform the training in my Colab and build and run the C++ project for inference.