Skip to content

Himanshu-1988/massachusetts-roads-image-segmentation

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

40 Commits
 
 
 
 
 
 

Repository files navigation

massachusetts-roads-image-segmentation

Problem : to find features from the image dataset

Approach :

  • 1.Downloading data from Kaggle

for downloading data from Kaggle either we can download it from the website or use kaggle.json to connect to Kaggle. once connected we can download the zip file

  • 2.unzipping the data

once we have the zip files we have to extract our data so that we can use it further

  • 3.Checking some sample images

after extraction, we will take a random sample to check the images by visualizing them

  • 4.iterate over the data

now we have got the images we will iterate through them and do the pre-processing steps

  • 5.Data Preprocessing - we will be changing image to array then we will be resizing them all to the same size for that we will be using tf dataset

  • 6.Data Augmentation and Train_Test split

once we were done with the pre-processing we will do train test split and do data augmentation using generate function which we have created

  • 7.Using UNet for image segmentation with dice loss as loss and metrics as IOU

we will use UNET to segment the image and extract its feature. we will be using dice loss,IOU,precision and accuracy

  • 8.while model training use callbacks for checkpointing and early stopping

we will train our model and then plot the losses and learning curve. we could train for more epochs based upon time and GPU confirmation

  • 9.Evaluation we will perform the evaluation on the validation images provided to us

    • evaluation results : loss: 0.1131 - iou: 0.4071 - mean_io_u: 0.4713 - dice_coef: 0.5711 - precision: 0.8596
  • 10.Visualize results on the validation image

-----Image--------------------------------Mask------------------------Predicted-------

Screenshot

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published