Authors: Àlex Vicente, Àlex Palomo, Joan Sintes and Roger Marí (2017)
Week 1:
Fill the 'test_2017' and 'train_2017' folders with the test and train splits.
Go to the 'week1' folder and execute the 'main.m' script with Matlab.
The 'main.m' script contains the code for tasks 1, 2, 3 and 4.
The different strategies that we used to solve each task are commented in detail in the scripts.
In order to generate the masks for the test split, execute the 'test.m' script with Matlab.
The output binary masks will be stored as png files in the following directory: test_2017/test/mask.
Week 2:
Fill the 'test_2017' and 'train_2017' folders with the test and train splits.
Go to the 'week2' folder and execute the 'task2.m' script to check task 2.
The implementation of the morphological operators required for task 1 can be found in the /morphological_operators folder.
Execute the 'main.m' script to check tasks 3 and 4.
The different strategies that we used to solve each task are commented in detail in the scripts.
In order to generate the masks for the test split, execute the 'test_methodX.m' scripts.
test_method1.m: HCbCr color segmentation with hand picked thresholds + post-processing.
test_method2.m: Histogram back-projection color segmentation (with Cb and Cr channels from YCbCr space) + post-processing.
test_method3.m: Histogram back-projection color segmentation (with Hue and Saturation channels from HSV space) + post-processing.
Week 3:
Fill the 'test_2017' and 'train_2017' folders with the test and train splits.
Go to the 'week3' folder and execute the 'main.m' script to test the code with the validation set.
Task 1 is implemented in the /Compute_Mask_Functions/CCL_filtering.m function.
Task 2 is implemented in the /Compute_Mask_Functions/SLW_filtering_basic.m function.
The multiple detections problem is adressed witht the /Compute_Mask_Functions/merge_windows.m function.
Task 3 is implemented in the /Compute_Mask_Functions/SLW_integral_image.m function.
Task 4 is implemented in the PerformanceAccumulationWindow.m and PerformanceEvaluationWindow.m functions.
The different strategies that we used to solve each task are commented in detail in the scripts.
In order to generate the masks and window candidates for the test split, execute the 'test_methodX.m' scripts.
test_method1.m: H-S color seg., + Hole fill. + Opening + Connected Components (CC) filtering based on form factor and size.
test_method2.m: H-S color seg., + Hole fill. + Opening + Sliding window to filter CC based on filling ratio.
Week 4:
Fill the 'test_2017' and 'train_2017' folders with the test and train splits.
Go to the 'week4' folder and execute the 'main.m' script to test the code with the validation set.
Task 1 is implemented in the /Compute_Mask_Functions/template_matching_corr.m function.
Task 2 is implemented in the /Compute_Mask_Functions/template_matching_DT.m function.
The different strategies that we used to solve each task are commented in detail in the scripts.
In order to generate the masks and window candidates for the test split, execute the 'test_methodX.m' scripts.
test_method1.m: H-S color seg., + Hole fill. + Opening + CC filt. (form factor, size) + Template Matching (correlation)
test_method2.m: H-S color seg., + Hole fill. + Opening + CC filt. (form factor, size) + Template Matching (distance transform)
Week 5:
Fill the 'test_2017' and 'train_2017' folders with the test and train splits.
Fill the mcg-2.0/pre-trained folder after donwloading the MCG code from the link below.
https://www2.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/mcg/
Go to the 'week5' folder and execute the 'main.m' script to test the code with the validation set.
Task 1 and 2 are implemented in the /Compute_Mask_Functions/scg_segmentation.m function.
The different strategies that we used to solve each task are commented in detail in the scripts.
In order to generate the masks and window candidates for the test split, execute the 'test_methodX.m' scripts.
test_method1.m: SCG seg., + CC filt. (form factor, size) + Histogram Comparison + Template Matching (distance transform)
test_method2.m: SCG seg., + CC filt. (form factor, size) + Histogram Comparison + Hough Transfrom