Closed rishabh063 closed 5 months ago
Aha, thx for the continual attention :) I did a similar thing in InSPyReNet to train the model with massive datasets. In my setting, the massive training set is [DIS-TR, DIS-TEs] + [HRSOD-TR, HRSOD-TE] + [UHRSD-TR, UHRSD-TE] + [HRS10K-TR, HRS10K-TE], and the DIS-VD is left as validation set. You can find it in config.py.
Thanks for the training , how are the results when compared to closed source solutions ?
i made a quick colab file for people to try this repo in a easier way for single image inference ( i have written this 5-10 times but always loose it edit the same colab like a noob)
attaching the link : - https://colab.research.google.com/drive/1aCBF9THgyHHxJwoIYfkYfHOjv6m0y1A8?usp=sharing.
Do make a copy and add to read me ( if you like it ) . People new to ml (like me ) will like this kind of inference
Good! Thanks. I'll add the link to it in the README. About the model trained on massive data, it's not better than the BiRefNet-DIS on DIS-VD. But I think it should work better on the wild images.
hey saw i new checkpoint in stuff what is it?