generation.md

Detection models

In this section, we provide the code to test Convolution-based and Transformer-based models on the CoMix dataset. You can find the weights here.

Create a subfolder weights inside benchmarks. Now inside weights create three subdirectories dass, faster-rcnn and yolov8.

Regarding dass and faster-rcnn put all the .pth files in the corresponding folders. Instead, in yolov8 put ONLY yolov8x-best.pt, yolov8x-c100.pt, yolov8x-m109.pt and yolov8x-mix.pt.

All predictions will be stored in data/predicts.coco. For semplicity, we are going to create a conda environment for all the scripts that are needed and in order to install every dependency.

DASS

First of all under CoMix create a folder modulesand inside of it clone this repository.

Then, create the conda environment and activate it (remeber to deactivate the conda environment you were using before!):

$ conda create -n dass python=3.8
$ conda activate dass

Now, install dependencies:

$ pip install numpy
$ pip install chainercv
$ pip install -e .
$ pip install loguru
$ pip install xmltodict
$ pip install requests-doh

Run the following command for every dataset and for every weights file:

$ python benchmarks/detections/dass.py -n dataset_name -s split_name -pd weights_file

Where: dataset_name -> choices = ['eBDtheque', 'DCM', 'comics', 'popmanga']

split_name -> choices = ['val', 'test']

weights_file -> choices =['m109', 'dcm', 'mixdata']

Faster R-CNN

Create the conda environment and activate it (remeber to deactivate the conda environment you were using before!):

$ conda create -n faster python=3.8
$ conda activate faster

Now, install dependencies:

$ pip install -e .

Run the following command:

$ python benchmarks/detections/faster_rcnn.py -n dataset_name -s split_name -wn weights_file

Where: dataset_name -> choices = ['eBDtheque', 'DCM', 'comics', 'popmanga']

split_name -> choices = ['val', 'test', 'all']

weights_file -> choices =['faster_rcnn-c100-best-10052024_092536.pth', 'faster_rcnn-c100-last-10052024_092536.pth', 'faster_rcnn-m109-best-10052024_094048.pth', 'faster_rcnn-mix-best-10052024_112437.pth']

YOLOv8

Create the conda environment and activate it (remeber to deactivate the conda environment you were using before!):

$ conda create -n yolo python=3.8
$ conda activate yolo

Now, install dependencies:

$ pip install -e .

Run the following command:

$ python benchmarks/detections/yolov8.py -n dataset_name -s split_name -wn weights_file

Where: dataset_name -> choices = ['eBDtheque', 'DCM', 'comics', 'popmanga']

split_name -> choices = ['val', 'test', 'all']

weights_file -> choices =['yolov8x-best.pt', 'yolov8x-c100.pt', 'yolov8x-m109.pt', 'yolov8x-mix.pt']

GroundingDINO

Create the conda environment and activate it (remeber to deactivate the conda environment you were using before!):

$ conda create -n dino python=3.8
$ conda activate dino

Now, install dependencies:

$ pip install -e .
$ pip install requests-doh

Run the following command:

$ python benchmarks/detections/groundingdino.py -n dataset_name -s split_name

Where: dataset_name -> choices = ['eBDtheque', 'DCM', 'comics', 'popmanga']

split_name -> choices = ['all', 'val', 'test']

Magi

Create the conda environment and activate it (remeber to deactivate the conda environment you were using before!):

$ conda create -n magi python=3.8
$ conda activate magi

Now, install dependencies:

$ pip install -e .
$ pip install requests-doh
$ pip install transformers==4.45.2

Run the following command:

$ python benchmarks/detections/magi.py -n dataset_name -s split_name

Where: dataset_name -> choices = ['eBDtheque', 'DCM', 'comics', 'popmanga']

split_name -> choices = ['val', 'test']

SSD

To be added.