From 71fb11bb934c1119427006dd01c43c2c8372bfbb Mon Sep 17 00:00:00 2001 From: MathieuNlp Date: Fri, 6 Oct 2023 15:54:52 +0200 Subject: [PATCH] fix --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index 6cfb407..15f8055 100644 --- a/README.md +++ b/README.md @@ -213,7 +213,7 @@ We can see some good segmentation like in demo 1 or demo 2 but it becomes more d # Conclusion/Discussion Using adapters, I partially succeded the task of segmenting rings. In the demo examples we saw that the model still struggles to caputre all the rings. This could be because the training dataset is small (only 8 images). -To improve the model, we can train on higher number of epochs, add model checkpoint and early stopping with a validation set to prevent from overfitting. In addition, doing data augmentation could be interesting to strengthen the model generalization. +To improve the model (if enough memory), we could train on higher number of epochs, add model checkpoint and early stopping with a validation set to prevent from overfitting. In addition, doing data augmentation could be interesting to strengthen the model generalization. Another interesting idea could be to apply LoRA to other blocks of SAM like the mask decoder which is smaller than the image encoder. This could save training time and potentially show better results.