search

tryolabs / luminoth

Deep Learning toolkit for Computer Vision.
https://tryolabs.com
BSD 3-Clause "New" or "Revised" License
2.41k stars 399 forks source link

TODO guidelines on how to write your own conversion tool #112

Open singhanj13 opened 6 years ago

singhanj13 commented 6 years ago

I could not find the guidelines for writing my own conversion tool on a new dataset.

clicit commented 6 years ago

I was also searching for some guidelines but couldn't find any. But maybe this branch helps to get started: https://github.com/tryolabs/luminoth/tree/custom-dataset

vierja commented 6 years ago

Hi @singhanj13 @clicit, we are currently rewriting the conversion tool in https://github.com/tryolabs/luminoth/tree/transform-rewrite and we'll add docs and sample code on how to write custom readers as well.

May I ask in what way do you have the data stored? We definitely want to have some basic custom readers by default.

clicit commented 6 years ago

Hi @vierja , my dataset comes from a Keras SSD. Beside two image directories (train, val) the labels are stored in CSV files with format:

img_name, x_min, x_max, y_min, y_max, class_id

Would be great if you could give some advice!

many thanks!

vierja commented 6 years ago

@clicit I've added a basic CSV reader for what you were describing (it's in bd0df9cd7aef7426696966608cbf6eb0f57d859d, still not in master). I would love to hear more about your use case and how we can improve Luminoth for broader use cases, if you are interested in telling me a bit more you can reach me at jrey at tryolabs com.

singhanj13 commented 6 years ago

Hi @vierja , my dataset is in a similar format as Pascal VOC. I have JPEG images, image sets containing lists of training, validation and test images and bounding box annotations in XML format.

BHunter2889 commented 6 years ago

Hello @vierja - I realize this is an old issue but since you haven't closed it I assume you're still adding support. Is there any shot at transforming .mat annotations? Would be a huge help if so. Thanks!