Open anto-madaro opened 4 years ago
Hi, you don't need to modify sample/trains/detectors, just add a new file for your own dataset under CenterNet/src/lib/datasets/dataset/
, import it in CenterNet/src/lib/datasets/dataset_factory.py
and run.
Thanks for you advices, I managed to train Centernet on my own dataset, creating my model in pth format with a size of 2 GB after a training phase on almost 8.000 images. I tested my model with your test.py and it produced a json dictionary with format: category_id -> [annotations] where each annotation is composed of a bbox with xy, wigth and height and a score. I tried manually to draw boxes, found at exp/ctdet/results.json on test images and it works quite fine. Now, I want to produce an only detection so at first I run your demo with my model on a small , new video but I think there are some errors with categories: my model was trained for 1 category(basketball player) but demo predicts only wrong categories (cows, microwaves,ecc... or nothing).
Have I to modify something for running demo correctly?
Hi, this is my first machine learning project and I'm confused.
How do you add your own dataset to CenterNet/src/lib/datasets/dataset/? My dataset is a bunch of images. I know I'm not supposed to add that to CenterNet/src/lib/datasets/dataset/, so what am I supposed to add? What should the file I add contain? Annotations? Paths? I'm so confused.
Thnak you for your help
Seconding the above question
@howdidthatwork @Zireael07 This is my first time working on a machine learning project as well. Were you able to resolve the issue? If yes, can you please share the details, it might really be helpful for someone who's starting out. Thanks.
Hi, I'm interested in training your detector with hourglass but using my dataset. In particoular is composed by an half million basket images, so with just a 'people' cathegory. Trying to recap all passages:
Is that alright or I'm missing something? Thanks for your time and your amazing job.