The aim of this project is to create augmented labelled images from already existing ones. Specifically, two datasets were examined: the first includes images taken from a cataract operation, while the second involves images depicting vocal folds. The proposed approach generates the augmented labels by recombining various fragments from the original labels, through a logical program implemented in Answer Set Programming (ASP). The consistency of the resulting augmented image is ensured by the ASP program itself. The ASP programs responsible for generating augmented labels regarding cataract and vocal folds can be found in the cataract and vocalfolds folders, respectively.
In order to properly run the ASP programs, one must dispose of:
- DLV version 2.1.0 (static executables with support to Python external atoms is mandatory) - can be downloaded form the following link: https://dlv.demacs.unical.it/home
- Python 2.7 (this specific version of the Python interpreter is required in order to run the DLV solver)
- Linux x86_64
The program augmented_label_generator_cataract.asp takes in input some matrices of 960x540 pixels, represented by facts in the following form:
cell(R,C,CLASS)
where CLASS is the class of the cell at row R and column C.
For each matrix, it must exist an atom of the form class(FILENAME), where:
classis the name of the class this matrix representsFILENAMEis the name of the image file these facts represent
This program gives in output some facts in the form new_cell(R,C,CLASS), which represent the new matrix (thus, the new augmented image).
From the folder cataract run the following command:
<path_to_dlv2_solver> augmented_label_generator_cataract.asp external_atoms.py <path_to_input_file_1> <path_to_input_file_2> <path_to_input_file_n> --silent --filter=new_cell/3 > output.txt
where:
<path_to_dlv2_solver>is the path to the DLV2 solver<path_to_input_file_x>is a file containing all or some of the input factsoutput.txtis the file containing the output (note that, if the provided input violates one or more constraints, then output.txt will containINCOHERENT)
The program augmented_label_generator_vocalfolds.asp takes in input some matrices of 512x512 pixels, represented by facts in the following form:
cell(R,C,CLASS)
where CLASS is the class of the cell at row R and column C.
For each matrix, it must exist an atom of the form class(FILENAME), where:
classis the name of the class this matrix representsFILENAMEis the name of the image file these facts represent
This program gives in output some facts in the form new_cell(R,C,CLASS), which represent the new matrix (thus, the new augmented image).
From the folder vocalfolds run the following command:
<path_to_dlv2_solver> augmented_label_generator_vocalfolds.asp external_atoms.py <input_group> <input_background> <input_pathology> <input_intubation> <input_surgical_tool> --silent --filter=new_cell/3 > output.txt
where:
<path_to_dlv2_solver>is the path to the DLV2 solver<input_group>(mandatory) is the path to a txt file containing only one fact of the form:group(X), whereXis a number from1to5. This refers to the group of the image one wants to generate.<input_background>(mandatory) is the path to atxtfile containing the facts which represent the background<input_pathology>(optional, depends on group) is the path to a txt file containing the facts which represent the pathology<input_intubation>(optional, depends on group) is the path to a txt file containing the facts rhich represent the intubation<input_surgical_tool>(optional, depends on group) is the path to a txt file containing the facts rhich represent the surgical tooloutput.txtis the file containing the output (note that, if the provided input violates one or more constraints, then output.txt will containINCOHERENT)