2D_Object_Detection_Gr48

Use of Retinanet for cityscapes - Project DLAV 2023

Folder Structure

Download the Cityscapes Dataset and organize the files in the following structure. Create an empty annotations directory. The conversion_seg_to_bbox has to contain the following (the Cityscape dataset gtFine and leftImg8bit can be obtain online by downloading a zip):

data/
└── cityscapes
    ├── annotations
    ├── gtFine
    │   ├── test
    │   ├── train
    │   └── val
    └── leftImg8bit
        ├── test
        ├── train
        └── val
main.py
inspect_coco.py
requirements1.txt

You should run the code provided from a iypnb create in the same repo as Retinanet see below (here called main.ipynb)

Retinanet/
README.md
main.iypnb

Installation of the requirements for the data segmentation conversion to box

!pip install -r Retinanet/conv/requirements1.txt 

Run the bbox on the Cityscapes

To run the conversion execute the following

!python Retinanet/conv/main.py --datadir Retinanet/conv/data/cityscapes --outdir Retinanet/conv/data/cityscapes/annotations

In order to run the visualization of the Cityscapes dataset with boxes you may run

!python Retinanet/conv/vizubbox.py

Input 1 - Segmentation

Output 1 - Bbox

Installation of the requirements for the model

!pip install -r Retinanet/requirements2.txt 

Run the Retinanet model

To run the model execute (you might have to uncomment the model)

!python Retinanet/main_model.py

In order to run the visualization of Cityscapes prediction please launch run the command:

!python Retinanet/vizualization.py

note: Due to computation issues the weights that are used have been computed on the initial dateset

Output 2