[WIP] Analyse false predictions

[Trion129]

Fixes #109.

TODO:

• Load all images and make predictions
• Compare to extract false predictions
• Create a GUI to display false predicted images in sequence
• Create a next and previous button to go back and forth
• also bind keyboard arrows.
• Be able to pop the image label if the images are too noisy to learn

[marco-c]

@Trion129 are you still interested in working on this?

[Trion129]

Yep, I just have some quizzes till 18th at college. After that I'll work on this :)

[marco-c]

ping @Trion129

[Trion129]

Hi @marco-c, I was busy last month at college. I had made some progress, but I was having issue with placing the "next" button properly, it gets overlapped under the images. I'll push the code anyways, you might have a fix.

[Trion129]

On adding the image, it covers the next button. Even though they are in separate grids.

[Trion129]

Never-mind, it started to work when I put the button inside the image-view itself. Weird tkinter...

[Trion129]

@marco-c For the popping of images, as the pics are spread over many csv files, I think it is better to generate a new list "false_pairs.csv" where the image names will be appended. This list can be used by pretrain and train scripts to remove false pairs from data. What's your opinion?

[marco-c]

@marco-c For the popping of images, as the pics are spread over many csv files, I think it is better to generate a new list "false_pairs.csv" where the image names will be appended. This list can be used by pretrain and train scripts to remove false pairs from data. What's your opinion?

I think we shouldn't pop those images, let's just implement the analysis of false predictions.