AutoGradeSystem

Here are the codes of the project AutoGradeSystem. The objective is to automatically grade PCV/AIA/DIP/ exams.

Getting started

Python & Required Libraries

The whole code are written in Python3 and a couple of libraries are used in the code. Make sure the following libraries are installed.

python >= 3.6.7
numpy >= 1.16.4
pandas >= 0.24.2
opencv-python >= 4.1.0
scipy >= 1.3.0
sklearn >= 0.21.2

How to use

Just run the script main.py located on the root directory without the any parameter to test it on the our collected scans. If no error occurs, a folder called './all_results/' will be created, where the results will be stored. Several parameters within the main.py need to be clarified here.

digit_recognize_on (line 29): if True, then the student IDs will be recognized automatically. Otherwise the student IDs are supposed to be given with a .csv file. For instance an example file called student_ids_example.csv is provided in folder '.inputs/'. The standard solution file 'solution_example.xlsx' is also located in that folder . By default it is set to True. The user will be firstly required to select the Region of the Interest (ROI) that contains the digits string by dragging a rectangular. Moreover, user can left click on the image to reselect the region and press ENTER to confirm your selection. More details about the digits recognition will be showed in following section.

Select ROI
semi_mode_on (line 27): if True, human intervention is required for each answer sheet. By default it is set to False. When human intervention is required, normal mode is activated and you can left click the cell to mark as cross and right click to mark the cell as non-cross. You can also press e to enter edit mode, which allows to map the rows and press e again to quit that mode. By default it is set to False

Enter edit mode

Digits recognization

Thanks for putting back the deadline so that we can have time for the implement of the digits recognition module. we use the pca and svm to predict the single digit. the pre-trained modules are stored in folder './trained_models/'. The data set for training the modules is MNIST. Since its the US written style of digits is different from german style. Its quite often to see that in our test scans digit 1 is miss-classified as 7 and digit 7 is miss-classified into 2. You can check those sheets in folder './example_results/CoverSheets/'. An example taken from that folder looks like below.

Paper source code

location: './Latex/'

Contributions

Paper: Yang Xu 60% Lei Jiao 40%

codes: Yang Xu 50% Lei Jiao 50%

LeiDoubi / AutoGradingSystem