Weather Image Recognition

Erbo Shan

This project includes four .py files

All four .py files are used for training the data set.

• WebCam.py: make predictions without relabling and normalization
• WebCamNormalized.py: make predictions with normalization
• Relable.py: relabl and normalize
• predict.py: takes images to predict the weather and time by using greyscale method and colourful image method.

  python3 predict.py katkam-20170203150000.jpg 
  

If you are interested in using Tensorflow in this project:

• Download retrained_graph.pb

• Download tf_files.zip and uncompress it

• Open terminal

• Get testing script

  curl -L https://goo.gl/3lTKZs > label_image.py

• This following command is used for testing:

  python label_image.py tf_files/Cloudy/Cloudy329.jpg

• Something like this will shown below:

	mostlycloudy (score = 0.34056)
	clear (score = 0.14641)
	mainly clear (score = 0.14396)
	cloudy (score = 0.12643)
	rain showers (score = 0.08014)
	drizzle (score = 0.06460)
	moderate rain (score = 0.03207)
	rain (score = 0.03091)
	fog (score = 0.01696)
	rain fog (score = 0.01162)
	snow (score = 0.00633)

• We have already changed the name of image by its label (weather)[if there is no related weather from CSV file, we just ignore them], and classified them into different labeled (weather)folders.

eg. if we can find a specific weather for katkam-20160605080000.jpg (Let's say it is Rain). 
The name of this images will be modified to Rainxx.jpg

• If you want to rename each image. The function below will help (uncommend it in WebCam.py ):+1: # renaming(FileName, Classifier)

If you want to retrain with your own data set

• Tensorflow is required but Docker is more strongly recommended

• Work under the same directory as the tf_files

• Download traing script with this following command

curl -O https://raw.githubusercontent.com/tensorflow/tensorflow/r1.1/tensorflow/examples/image_retraining/retrain.py

• Start image retraining with this command and a retrained_graph.pb file will be generated

python retrain.py \
  --bottleneck_dir=bottlenecks \
  --how_many_training_steps=500 \
  --model_dir=inception \
  --summaries_dir=training_summaries/basic \
  --output_graph=retrained_graph.pb \
  --output_labels=retrained_labels.txt \
  --image_dir=tf_files

Stuffs you may want to use for this(References):

• Tensorflow for trainning data
• Dcoker for training data on container
• References