search

Harry-KIT / Image-Classification-on-small-datasets-in-Pytorch

This repo is special for those who want to start learning computer vision related tasks such as image classification.
36 stars 7 forks source link

Dataset size #3

Open sila-a0 opened 1 year ago

sila-a0 commented 1 year ago

Hello, How many images from each class did you use while training? Thank you,

Harry-KIT commented 1 year ago

Hi @usertttwm. Thank you for your interest in our work. We used 376 images for each class.

Harry-KIT commented 1 year ago

Hi @usertttwm I hope the response I provided addressed your query. If you're satisfied and no longer require further assistance, kindly let me know so that I can proceed to close the issue. If you found the repository helpful, I would greatly appreciate it if you could consider giving it a star. Thank you!

sila-a0 commented 1 year ago

Hi @Harry-KIT , I don't know which model i should use in my own dataset. I have two classes. Currently there are 200 images from each class.(but i will increase to 600 from each class, I am collecting data) https://keras.io/examples/vision/image_classification_efficientnet_fine_tuning/ I tried like here, but i am getting some errors What is your advice to me? Thank you very much, I gave a star.

Harry-KIT commented 1 year ago

Hi @usertttwm, Could you please provide more details about the specific errors you mentioned? If you're encountering challenges with your dataset, utilizing augmentation methods might be helpful. If your issue is related to something else, kindly let me know. In any case, I've included a link below that you might find useful for augmentation techniques. https://github.com/DebeshJha/UNet-insturment-segmentation/blob/master/data_aug.py

sila-a0 commented 1 year ago

Hi @Harry-KIT I followed here : https://github.com/vatsaldpatel/EfficientNet-Transfer-Learning/blob/master/transfer_learning_EfficientNet.py this is mine: https://github.com/usertttwm/EfficientNetImageClassification/blob/main/dinnerTrayBreakfastTrayImageClassification.ipynb additionally I used callback for best model. and I printed the classification report. the result is always 1.00 overfitting? Do you have an idea ? Thank you very much

Harry-KIT commented 1 year ago

Hi @usertttwm , have you tried to use my repo? If yes, please let me know about the results?

sila-a0 commented 1 year ago

@Harry-KIT I haven't tried yet. I'll try and let you know, Thanks a lot