File size: 4,514 Bytes
45de7a1 c5bd7aa 422eed8 c5bd7aa 422eed8 c5bd7aa 422eed8 c5bd7aa 45de7a1 c5bd7aa 422eed8 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 |
---
language: en
tags:
- image-classification
- CNN
- Convolution Neural Entwork
- Nueral Network
- Trash
metrics:
- name: train-accuracy
value: 91%
- name: test-accuracy
value: 55%
pipeline:
- image-classification
libraries:
- name: torch
version: 1.9.0
- name: torchvision
version: 0.10.0
- name: numpy
version: 1.21.0
---
## Trash Classification CNN Model
### About
This project is a convolutional neural network (CNN) model developed for the purpose of classifying different types of trash items.
The CNN model in this project utilizes the TinyVGG architecture, a compact version of the popular VGG neural network architecture. The model is trained to classify trash items into the following subcategories:
- Cardboard
- Food Organics
- Glass
- Metal
- Miscellaneous Trash
- Paper
- Plastic
- Textile Trash
- Vegetation
In total, there are 9 categories into which the trash items are classified.
For more details about the CNN architecture used in this project, you can refer to the [CNN Explainer](https://poloclub.github.io/cnn-explainer/) website.
### Info
Only 30% of the data from the Real Trash Dataset has been used and divided into an 80%-20% split of Train and Test.
The Huggingface Repository contains 7 files found in the `files and versions` tab:
1. **data_setup.py**: This file contains functions for setting up the data into datasets using ImageFolder and then turning it into batches using DataLoader. It also returns the names of the classes.
2. **model_builder.py**: This file contains a class which subclasses nn.Module and replicates the TinyVGG CNN model architecture with a few modifications here and there.
3. **engine.py**: This file contains three functions: `train_step`, `test_step`, and `train`. The previous two are used to train and test the model, respectively, and the last one integrates both to train the model.
4. **plotting.py**: This file contains functions to plot metrics like loss and accuracy using `plot_metrics`, and it also has a function `plot_confusion_Matrix` to plot the confusion matrix.
5. **predict.py**: This file can be run with `--image` and `--model_path` arguments to get the prediction of the model on the specified image path.
6. **utils.py**: This file contains functions to save the model in a specific folder with a changeable name.
7. **train.py**: This script uses all the files except `predict.py` and can take argument flags to change hyperparameters. It can be run with the following arguments:
```
python train.py --train_dir TRAIN_DIR --test_dir TEST_DIR --learning_rate LEARNING_RATE --batch_size BATCH_SIZE --num_epochs NUM_EPOCHS
```
Additionally, it is device agnostic, meaning it automatically utilizes available resources regardless of the specific device used.
Additionally, the repository contains 2 folders:
- **data**: This stores the data and has subdirectories train and test.
- **models**: This stores the model saved by utils.py.
- **samples**: This has 10 pictures, you can use for testing the model using `predict.py`.
## Model Overview
This model is designed for image classification tasks. It requires input images of size 112x112 pixels. Containing 2 blocks with 2 convulutional layers and then a flattner with a classfier.
The architecture looks like :
```python
TrashClassificationCNNModel(
(block_1): Sequential(
(0): Conv2d(3, 15, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): Conv2d(15, 15, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU()
(4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(block_2): Sequential(
(0): Conv2d(15, 15, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): Conv2d(15, 15, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU()
(4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(classifier): Sequential(
(0): Flatten(start_dim=1, end_dim=-1)
(1): Linear(in_features=11760, out_features=9, bias=True)
)
)
```
## Dataset Overview
The dataset used containes images of multiple waste items with multiple classes named RealWaste. It has 4752 samples.
- Source: [Click here](https://archive.ics.uci.edu/dataset/908/realwaste)
- Citation: Single,Sam, Iranmanesh,Saeid, and Raad,Raad. (2023). RealWaste. UCI Machine Learning Repository. https://doi.org/10.24432/C5SS4G.
## Discliamer
The model mught give inaccurate or worng results BEWARE. |