Incorrect predictions

[sasinghi]

After training the model (for a short time) I see an accuracy of 80%. However, when I test it, the result is as shown for all test images. Can someone help with this? Would be great if someone could share the pre-trained weights.

[ajithvcoder]

model.fit_generator(myGene,steps_per_epoch=2000,epochs=5,callbacks=[model_checkpoint])

in above line change the steps_per_epoch to 300 or 400 or 500 some thing should work

[enggrey]

In case you or anyone else is still having issues with this, in the testGenerator function change this line
img = img / 255
for this
img = img / 255.

[sasinghi]

@ajithvallabai that was it. Training longer helped.
Thanks!

[duanmuji]

I modified img.255, but problems persisted.

[openXG]

model.fit_generator(myGene,steps_per_epoch=2000,epochs=5,callbacks=[model_checkpoint])

in above line change the steps_per_epoch to 300 or 400 or 500 some thing should work

I set steps_per_epoch=500, but my test images are still all gray...
should I set it more larger than 500? by the way, I just use CPU.

[JianyingLi]

model.fit_generator(myGene,steps_per_epoch=2000,epochs=5,callbacks=[model_checkpoint])
in above line change the steps_per_epoch to 300 or 400 or 500 some thing should work

I set steps_per_epoch=500, but my test images are still all gray...
should I set it more larger than 500? by the way, I just use CPU.

i think epochs should be increased rather than steps_per_epoch to improve the acc . Another place to check is the function of adjustData.
def adjustData(img,mask,flag_multi_class,num_class):
# here#
elif(np.max(img) > 1):
img = img / 255
mask = mask /255
mask[mask > 0.5] = 1
mask[mask <= 0.5] = 0
return (img,mask)
This code works for py3, if you use py2, modify '255' to '255.'; Last, you need to check the threshold '0.5' in order to make sure the returned mask is a binary(0/1) mask.

[lihao6666]

get it

[jizhang02]

@JianyingLi Hello,
I change cross entropy loss into dice, but the result become all white, could you help me?
def dice_coef(y_true, y_pred):
smooth = 1
y_true_f = K.flatten(y_true)
y_pred_f = K.flatten(y_pred)
intersection = K.sum(y_true_f * y_pred_f)
return (2. * intersection +smooth) / (K.sum(y_true_f) + K.sum(y_pred_f) +smooth)

def dice_coef_loss(y_true, y_pred):
print("dice loss")
return 1-dice_coef(y_true, y_pred)

model.compile(optimizer = Adam(lr = 1e-5), loss = dice_coef_loss, metrics = ['accuracy'])