How to improve object detector?

Hello guys, I trained my yolov4 detector using 350 images and 50 epochs… but average precision for each object was not as good as I want… what can I do to increase the average precision?

Answer 1

[Image Analyst]

Train with more images and train for longer.

[Adrian Kleffler]

And do you know how to remove multiple images from table in matlab when i know their path?

[Image Analyst]

You could use contains or strcmpi to find rows that match the path you want to remove.

[Adrian Kleffler]

i have code like this to remove rows...

% Nahratie tabulky LabelData
load('LabelData.mat');

% Zadanie ciest k obrazkom, ktoré chceš vymazať
pathsToDelete = {'E:\ADRIAN\BAKALARKA\DATASET\vsetko2\panasonic_fullhd_01-090-045-202111091200.jpg', 'E:\ADRIAN\BAKALARKA\DATASET\vsetko3\fullhd-090-000-201901101000-01.png','E:\ADRIAN\BAKALARKA\DATASET\vsetko2\panasonic_fullhd_01-090-000-202109091400.jpg','E:\ADRIAN\BAKALARKA\DATASET\vsetko2\panasonic_fullhd_01-090-000-202110090800.jpg','E:\ADRIAN\BAKALARKA\DATASET\vsetko2\panasonic_fullhd_01-090-000-202103091300.jpg','E:\ADRIAN\BAKALARKA\DATASET\vsetko3\fullhd-090-000-201901102200-01.png'};

% Vymazanie riadkov s danými cestami
for i = length(pathsToDelete):-1:1
    idx = strcmp(LabelData.imageFilename, pathsToDelete{i});
    LabelData(idx, :) = [];
end

% Uloženie upravenej tabuľky
save('LabelData.mat', 'LabelData');

but i have 888 images and this code removes 777 images instead of only 5 images which i want to remove

[Image Analyst]

You keep forgetting to attach your .mat file. I can't try anything until you do. I'll check back later for it.