Accuracy cannot get more than 70%

Part 1 (2019)
1

Hi, I am new in deep learning. I need to train a network which is technically cell detection. cropped_jpg_55_199_0 cropped_png_55_199_0

I uploaded 2 images which are examples from my data set. Here I will also upload my program as well:data_gen_args = dict(rotation_range=0.2, width_shift_range=0.05, height_shift_range=0.05, shear_range=0.05, zoom_range=0.05, horizontal_flip=True, fill_mode='nearest') myGene = trainGenerator(2,'/Users/khand/OneDrive/Desktop/Thesis/Case_db/train_data','org','png',data_gen_args,save_to_dir = None) model = unet() model_checkpoint = ModelCheckpoint('unet_membrane.hdf5', monitor='loss',verbose=1, save_best_only=True) model.fit_generator(myGene,steps_per_epoch=100,epochs=10, callbacks=[model_checkpoint])

So if you have any suggestion how I can Improve my program in order to have higher accuracy. I would like to hear any ideas.
Thanks

2

Is it possible to post your data and existing code somewhere? This would help significantly with giving suggestions on how to improve your setup.

3


This is code.
And I have data consisted of 480 images I can increase it to 560, and data is in two folders and the images I uploaded are in examples from folders.

4

Since this is the fastai forums, I would recommend using fastai. :grinning:
The code you shared uses Keras.

5

I see, could you share a source for me from where I get a code and implement it to my dataset ?
Thanks

6

I would highly advice you to take the fastai course 2019 or at least watch lesson 3 which tackles the segmentation problem.

1 Like