Image-Classification-Transfer-Learning

Building ResNet152V2 Model for Image Classification with Small Dataset (95% accuracy)

Number of classes: 20 (Classes 0-19)

Classes = owl | galaxy | lightning | wine-bottle | t-shirt | waterfall | sword | school-bus | calculator | sheet-music | airplanes | lightbulb | skyscraper | mountain-bike | fireworks | computer-monitor | bear | grand-piano | kangaroo | laptop ]


Dataset Structure
Two folders:
Training: 1554 images
Test: 500 images

Images per class:

school-bus : 73
laptop : 100
t-shirt : 100
grand-piano : 70
waterfall : 70
galaxy : 56
mountain-bike : 57
sword : 77
wine-bottle : 76
owl : 95
fireworks : 75
calculator : 75
sheet-music : 59
lightbulb : 67
bear : 77
computer-monitor : 100
airplanes : 100
skyscraper : 70
lightning : 100
kangaroo : 57   

visualization of training data

Result

The accuracy of the training reached 99.5% in 50 epoch.
The accuracy of the test reached 95% that i submitted to kaggle.

CSV file for kaggle submission

predicted_class_indices=np.argmax(pred,axis=1)
labels = train_gen.class_indices
labels = dict((v,k) for k,v in labels.items())
predictions = [k for k in predicted_class_indices]

filenames=test_gen.filenames
FN=[]
for i in filenames:
  f = i[5:]
  FN.append(f)
 
results=pd.DataFrame({"Id":FN,
                      "Category":predictions})
results.to_csv("submission.csv",index=False)