6 minute readCreated on January 16, 2020Last modified on January 19, 2020

Creating a Custom Image Classifier using Turicreate to detect Smoke and Fire

For setting up Kaggle with Google Colab, please refer to my previous post

Dataset

Mounting Google Drive

import os
 from google.colab import drive
 drive.mount('/content/drive')
-
-
-

Downloading Dataset from Kaggle

os.environ['KAGGLE_CONFIG_DIR'] = "/content/drive/My Drive/"
+

Downloading Dataset from Kaggle

os.environ['KAGGLE_CONFIG_DIR'] = "/content/drive/My Drive/"
 !kaggle datasets download ashutosh69/fire-and-smoke-dataset
 !unzip "fire-and-smoke-dataset.zip"
-
-
-

Pre-Processing

!mkdir default smoke fire
-
-
-

!ls data/data/img_data/train/default/*.jpg
-
-
-

img_1002.jpg   img_20.jpg     img_519.jpg     img_604.jpg       img_80.jpg
+

Pre-Processing

!mkdir default smoke fire
+

!ls data/data/img_data/train/default/*.jpg
+

img_1002.jpg   img_20.jpg     img_519.jpg     img_604.jpg       img_80.jpg
 img_1003.jpg   img_21.jpg     img_51.jpg     img_60.jpg       img_8.jpg
 img_1007.jpg   img_22.jpg     img_520.jpg     img_61.jpg       img_900.jpg
 img_100.jpg    img_23.jpg     img_521.jpg    'img_62 (2).jpg'   img_920.jpg
@@ -47,49 +39,39 @@
 img_204.jpg    img_501.jpg    img_601.jpg     img_78.jpg
 img_205.jpg    img_502.jpg    img_602.jpg     img_79.jpg
 img_206.jpg    img_50.jpg     img_603.jpg     img_7.jpg
-
-
-

The image files are not actually JPEG, thus we first need to save them in the correct format for Turicreate

from PIL import Image
+

The image files are not actually JPEG, thus we first need to save them in the correct format for Turicreate

from PIL import Image
 import glob
 
 
 folders = ["default","smoke","fire"]
 for folder in folders:
   n = 1
-  for file in glob.glob("./data/data/img_data/train/" + folder + "/*.jpg"):
-    im = Image.open(file)
+  for file in glob.glob("./data/data/img_data/train/" + folder + "/*.jpg"):
+    im = Image.open(file)
     rgb_im = im.convert('RGB')
     rgb_im.save((folder + "/" + str(n) + ".jpg"), quality=100)
     n +=1 
-  for file in glob.glob("./data/data/img_data/train/" + folder + "/*.jpg"):
-    im = Image.open(file)
+  for file in glob.glob("./data/data/img_data/train/" + folder + "/*.jpg"):
+    im = Image.open(file)
     rgb_im = im.convert('RGB')
     rgb_im.save((folder + "/" + str(n) + ".jpg"), quality=100)
     n +=1
-
-
-

!mkdir train
+

!mkdir train
 !mv default ./train
 !mv smoke ./train
 !mv fire ./train
-
-
-

Making the Image Classifier

Making an SFrame

!pip install turicreate
-
-
-

import turicreate as tc
+

Making the Image Classifier

Making an SFrame

!pip install turicreate
+

import turicreate as tc
 import os
 
-data = tc.image_analysis.load_images("./train", with_path=True)
+data = tc.image_analysis.load_images("./train", with_path=True)
 
 data["label"] = data["path"].apply(lambda path: os.path.basename(os.path.dirname(path)))
 
-print(data)
+print(data)
 
 data.save('fire-smoke.sframe')
-
-
-

+-------------------------+------------------------+
+

+-------------------------+------------------------+
 |           path          |         image          |
 +-------------------------+------------------------+
 |  ./train/default/1.jpg  | Height: 224 Width: 224 |
@@ -123,9 +105,7 @@
 [2028 rows x 3 columns]
 Note: Only the head of the SFrame is printed.
 You can use print_rows(num_rows=m, num_columns=n) to print more rows and columns.
-
-
-

Making the Model

import turicreate as tc
+

Making the Model

import turicreate as tc
 
 # Load the data
 data =  tc.SFrame('fire-smoke.sframe')
@@ -141,16 +121,14 @@
 
 # Evaluate the model and print the results
 metrics = model.evaluate(test_data)
-print(metrics['accuracy'])
+print(metrics['accuracy'])
 
 # Save the model for later use in Turi Create
 model.save('fire-smoke.model')
 
 # Export for use in Core ML
 model.export_coreml('fire-smoke.mlmodel')
-
-
-

Performing feature extraction on resized images...
+

Performing feature extraction on resized images...
 Completed   64/1633
 Completed  128/1633
 Completed  192/1633
@@ -208,6 +186,4 @@
 Completed 384/395
 Completed 395/395
 0.9316455696202531
-
-
-

We just got an accuracy of 94% on Training Data and 97% on Validation Data!

Tagged with: