Bear Classifier

Using a pre-trained classifier to classify bears into ‘Teddy’, ‘Black’, and ‘Grizzly’ bears.

Bear Classifier

In this notebook, we will use a vision classifer to classfiy bear(teddy, black & grizzly) and deploy the model as web app using huggiface gradio space.

from fastai.vision.all import *

Data

The data is being imported from Kaggle

Dataset - anirudhg15/bears-fastai-2021

from google.colab import userdata
import os

os.environ["KAGGLE_KEY"] = userdata.get('KAGGLE_KEY')
os.environ["KAGGLE_USERNAME"] = userdata.get('KAGGLE_USERNAME')

!kaggle datasets download -d anirudhg15/bears-fastai-2021
! unzip "bears-fastai-2021"

from pathlib import Path

path = Path('bears')

bl_bear = (path/'black').ls().sorted()
bl_bear[1]

Path('bears/black/00000001.jpg')

Image.open(bl_bear[1])

DLS function fastai explaination

def get_dls(bs, size):
  dblock = DataBlock(blocks=(ImageBlock, CategoryBlock),
                     get_items=get_image_files,
                     splitter=RandomSplitter(),
                     get_y=parent_label,
                     item_tfms=Resize(size))

  return dblock.dataloaders(path, bs=bs)

dls = get_dls(64, 224)

dls.show_batch(nrows=1, ncols=3)

learn = cnn_learner(dls, resnet18, metrics=error_rate)
learn.fine_tune(4)

epoch	train_loss	valid_loss	error_rate	time
0	1.629151	0.250812	0.122807	00:21

epoch	train_loss	valid_loss	error_rate	time
0	0.429806	0.061032	0.017544	00:22
1	0.268742	0.013778	0.000000	00:21
2	0.186102	0.013141	0.000000	00:21
3	0.139788	0.016549	0.017544	00:20

Interpretting the model

interp = ClassificationInterpretation.from_learner(learn)
interp.plot_confusion_matrix()

interp.plot_top_losses(5, nrows=1, figsize=(17,4))

Prediction

learn.predict('/content/bears/black/00000017.jpg')

('black', tensor(0), tensor([9.9993e-01, 1.4458e-06, 7.3126e-05]))

Saving the Model

learn.export(fname='bear_model.pkl', pickle_module=pickle, pickle_protocol=2)

Reload Model

reload_model=load_learner('bear_model.pkl', cpu=True, pickle_module=pickle)
reload_model

<fastai.learner.Learner at 0x7bd9edf06890>

reload_model.predict('/content/bears/black/00000017.jpg')

('black', tensor(0), tensor([9.9993e-01, 1.4458e-06, 7.3126e-05]))