Ycb_Dataset_Generator

Installation

Commands to be executed in order:

1. git clone Ycb_Dataset_Generator
2. cd ./Ycb_Dataset_Generator
3. Setup and install Python 3.7.12 or Python 3.8.0 using Pyenv. CREATE and ACTIVATE a new python environment named 'YCB_generator' running Python 3.7.12 or Python 3.8.0. (This step is required to successfully meet all the requirements)
   • pyenv virtualenv 3.7.12 YCB_generator (Install python and create python virtual environment)
   • (Only for the initial setup) pyenv local YCB_generator (Activate python virtual environment)
   • pyenv versions (Check if the correct virtual environment with intended python version is active)
4. pip install -r ./Requirements/requirements.txt
5. python ./ycb_generate_cropped.py (Generates cropped images & masks of the YCB object(s))
6. mv ./models/ycb/* ./data (To use cropped objects from previous step in dataset generatation)
7. Open generator_for_yolov5.ipynb -> run all the cells => generates 100 train, 50 valid, 5 test 640x640 images with labels (yolo bounding box co-ordinates) by default.
8. Follow next instructions on Ycb_Yolov5_Trainer

Possible changes needed to run the notebook:

• Packages import configuration can be different different machines:

Custom dataset generation config:

• You can change the name of the custom dataset, number of images generated per split, andd add new split.

Resize dataset images:

• Recommended image sizes: 640x640 OR 416x416

Dataset Tips For Best Results: