PyTNet

A deep neural net Libraby Built on top of Pytorch

File Structure

1. Modles - Contains the dnn models

   • NewResnetModel.py - Dawnbench mark 2019 winners model
   • QuizModel.py - Dense Net
   • ResNetModel.py - Resnet
   • S7Model.py - custom model
   • MaskDepthModel - model that estimates both mask and depth
   • DepthModel - Model that estimates depth
   • MaskModel - Model that estimates mask

2. Dataset - contains data related modules

   • extract.py - Unzips the data set for monocular depth estimation and segmentation
   • MaskDepth.py - It brings the depth estimation andsegmentation to dataset format and applies the given transformations.
   • tinyimagenet - It downloads the tiny imagenet data, mix train-test, split into the given ratio and returns train and test set of type dataset.

3. Evaluation Metrics

   • Accuracy.py - Implements the dice score for evaluation of mask and depth.
   • loss.py - Implementation of different loss functions.

4. Results

   • showMnD - displays the predicted and target images of mask and depth.

5. Training

   • train_test_MnD.py - training for depth estimation and segmentation.
   • train_test.py - training for object recognisation.

6. Albumentation transforms - Used for Image Agumentations. It is from Albumentations library.

7. GradCam - Implements gradCam of the given images and specified layer of the model.

8. LrFinder - It finds the Lr of given range.

9. LR_Range_test - It finds the best Lr for One Cyce Policy

10. evaluate - It evaluates the final test accuracy, classwise accuracy, plots the given curves, gives misclassified inages and plots misclassified images,

11. show_images - Plots the given images of tensor for. Mainly used to visualise the train data.

12. train_test - Used to train the model.

13. train_test_loader - takes the train test data of type dataset, converts into data loader form, set the seed, check for the cuda availability.