Scene Classification in Images

Part 1 (2018)
1

I am trying to classify scenes in images I have about 17000 training images belonging to six classes, the classes are { 'buildings' -> 0,

'forest' -> 1,

`` 'glacier' -> 2,

`` 'mountain' -> 3,

`` 'sea' -> 4,

`` 'street' -> 5 }

I have finetuned Resnet50 on the dataset, initially I got accuracy of 93% without any data augmentation and after I tried data augmentation the accuracy decreased to 92% instead of increasing.
I used the following augmentation script
datagen = ImageDataGenerator(
rotation_range=45,
width_shift_range=0.4,
height_shift_range=0.4,
horizontal_flip=True,
fill_mode=“reflect”)
How can I increase the accuracy of this model? Here’s my accuracy and loss graph, the accuracy never goes beyond 0.93 despite augmentation of data.
accuracyloss

2

Hey you are on 5th position right on intel scene competition
I’m on 4th position username :- rishi34 accuracy :- 96.52936087

1 Like