Merge pull request #654 from Adhivp/American_sign_language_Enanchement

[Model Enhancement]: American Sign Language Detection Model
abhisheks008 · May 29, 2024 · 1c818a4 · 1c818a4
2 parents fc0ec4f + bb46650
commit 1c818a4
Show file tree

Hide file tree

Showing 13 changed files with 118 additions and 1 deletion.
diff --git a/American Sign Language Detection/README.md b/American Sign Language Detection/README.md
@@ -1,3 +1,4 @@
+# New enanchements has been made in the model , please check it below
 # American Sign Language Detection/https://github.com/World-of-ML/DL-Simplified/issues/312
 to predict correct sign language labels corresponding to their corresponding sign images
 
@@ -34,6 +35,13 @@ To implement InceptionV3, we start by loading the pre-trained model, which comes
 
 I will utilize the **VGG16** (Visual Geometry Group) architecture, which have deeper and complex structures. These models are renowned for their exceptional performance on various image recognition tasks. By leveraging the pre-trained weights of VGG, I can benefit from the learned features and fine-tune the network for image segmentation on the Lemon Quality Dataset.
 
+`new models implemented with new approach for enanchement`
+- ResNet101V2
+- ResNet50V2
+- MobileNetV3Large
+- MobileNetV3Small
+- InceptionV3
+- NASNetMobile
 
 **Accuracy Comparison**
 
@@ -46,6 +54,18 @@ I will utilize the **VGG16** (Visual Geometry Group) architecture, which have de
 
 Since the models' decent levels of accuracy(88% and above) means that most of their pictures will be almost havinG similar predicted labels with a small room for mistake, the anticipated labels for the sign image labels are as are visualised as follows.
 
+`new models accuracy`
+| Rank | Model Name       | Test Accuracy | Trained Model Size | Training Accuracy | Training Loss |
+|------|------------------|---------------|--------------------|-------------------|---------------|
+| 1    | MobileNetV3Small | 100.0%        | 19.1MB             | 96.97%            | 0.1574        |
+| 2    | NASNetMobile     | 100.0%        | 67.1MB             | 97.96%            | 0.1058        |
+| 3    | MobileNetV3Large | 100.0%        | 48.6MB             | 97.98%            | 0.1026        |
+| 4    | InceptionV3      | 100.0%        | 287.8MB            | 98.65%            | 0.0712        |
+| 5    | ResNet50V2       | 100.0%        | 308.6MB            | 98.67%            | 0.0625        |
+| 6    | ResNet101V2      | 100.0%        | 537.5MB            | 98.74%            | 0.0605        |
+
+- ranking based on Trained Model size
+
 **Throughout the project,** 
 
 I will preprocess the dataset by resizing the images and splitting it into training,validation and testing sets. For training, I will employ a loss function suitable for image segmentation, such as cross-entropy loss, and optimize the models using technique like  Adam optimization
@@ -60,8 +80,17 @@ I will evaluate their performance using appropriate metrics. Additionally, I wil
 
 ## after evaluation, `MobileNet` or `VGG16` model looks to be the best fit model in this case of American Sign Language Classification .
 
+## New models conclusian after enanchement
+- All models achieve a remarkable test accuracy of 100.0%, demonstrating their effectiveness in classification tasks.
+  - MobileNetV3Small stands out with a compact size of 19.1MB, offering high accuracy while minimizing resource usage, making it suitable for memory-constrained environments.
+  - NASNetMobile and MobileNetV3Large also deliver impressive accuracy with moderate model sizes, providing versatility in deployment scenarios.
+  - InceptionV3, ResNet50V2, and ResNet101V2, although larger in size, exhibit robust performance, with ResNet101V2 achieving the highest training accuracy of 98.74%.
 
 **Future Scope**
 
 This project will contribute to advancing the understanding and application of deep learning in the field of computer vision and could potentially find applications in sorting of sign languages in different classes.
 
+### ✒️ Improvements in this project is made by **Adhithyan VP**
+
+[![LinkedIn](https://img.shields.io/badge/LinkedIn-0077B5?logo=linkedin&logoColor=white)](https://www.linkedin.com/in/adhithyanvp)
+[![X (formerly Twitter) Follow](https://img.shields.io/twitter/follow/AdhiVp3)](https://x.com/AdhiVp3)
diff --git a/American Sign Language Detection/dataset/readme.md b/American Sign Language Detection/dataset/readme.md
@@ -1 +1,22 @@
 # link to the dataset - https://www.kaggle.com/datasets/ayuraj/asl-dataset
+
+# Models Enanchement
+## Link to the dataset
+- https://www.kaggle.com/datasets/kapillondhe/american-sign-language/data
+
+## Content
+The dataset contains coloured images of hand signs representing different American sign language alphabets.
+
+## Labels
+The dataset includes the following labels in alphabetical order:
+
+A, B, C, D, E, F, G, H, I, J, K, L, M, N, Nothing, O, P, Q, R, S, Space, T, U, V, W, X, Y, Z
+
+## Dataset Summary¶
+### Folder: train
+- Total images: 165670
+- Images per label: 5996 each
+
+### Folder: test
+- Total images: 112
+- Images per label: 4 each
diff --git a/American Sign Language Detection/images/Labels and Dataset Summary.png b/American Sign Language Detection/images/Labels and Dataset Summary.png
diff --git a/American Sign Language Detection/images/Sample images for each label.png b/American Sign Language Detection/images/Sample images for each label.png
diff --git a/...ican Sign Language Detection/images/True values and Predicted values (Test).png b/...ican Sign Language Detection/images/True values and Predicted values (Test).png
diff --git a/American Sign Language Detection/models/american-sign-language-inceptionv3.ipynb b/American Sign Language Detection/models/american-sign-language-inceptionv3.ipynb
diff --git a/American Sign Language Detection/models/american-sign-language-mobilenetv3Small.ipynb b/American Sign Language Detection/models/american-sign-language-mobilenetv3Small.ipynb
diff --git a/American Sign Language Detection/models/american-sign-language-mobilenetv3large.ipynb b/American Sign Language Detection/models/american-sign-language-mobilenetv3large.ipynb
diff --git a/American Sign Language Detection/models/american-sign-language-nasnetmobile.ipynb b/American Sign Language Detection/models/american-sign-language-nasnetmobile.ipynb
diff --git a/American Sign Language Detection/models/american-sign-language-resnet101v2.ipynb b/American Sign Language Detection/models/american-sign-language-resnet101v2.ipynb
diff --git a/American Sign Language Detection/models/american-sign-language-resnet50v2.ipynb b/American Sign Language Detection/models/american-sign-language-resnet50v2.ipynb
diff --git a/American Sign Language Detection/models/readme.md b/American Sign Language Detection/models/readme.md
@@ -1,6 +1,62 @@
- # American Sign Language Detection
+# American Sign Language Detection
 ![image](https://github.com/aditya0929/DL-Simplified/assets/127277877/c149f669-ed35-4751-87bd-b148495fafc4)
 
+## Enhanched models deatils
+### 🧮 **What I had done!**
+
+- I have imported various pretrained models from TensorFlow and added a softmax classification layer with 28 classifications.
+
+### 🚀 **Models Implemented**
+
+- ResNet101V2
+- ResNet50V2
+- MobileNetV3Large
+- MobileNetV3Small
+- InceptionV3
+- NASNetMobile
+
+### 📚 **Libraries Needed**
+
+- pandas
+- Pillow
+- numpy
+- tensorflow
+- matplotlib
+
+### 📊 **Exploratory Data Analysis Results**
+
+#### Folder: train
+- Total images: 165670
+- Images per label: 5996 each
+
+#### Folder: test
+- Total images: 112
+- Images per label: 4 each
+
+### 📈 **Performance of the Models based on the Accuracy Scores**
+
+| Rank | Model Name       | Test Accuracy | Trained Model Size | Training Accuracy | Training Loss |
+|------|------------------|---------------|--------------------|-------------------|---------------|
+| 1    | MobileNetV3Small | 100.0%        | 19.1MB             | 96.97%            | 0.1574        |
+| 2    | NASNetMobile     | 100.0%        | 67.1MB             | 97.96%            | 0.1058        |
+| 3    | MobileNetV3Large | 100.0%        | 48.6MB             | 97.98%            | 0.1026        |
+| 4    | InceptionV3      | 100.0%        | 287.8MB            | 98.65%            | 0.0712        |
+| 5    | ResNet50V2       | 100.0%        | 308.6MB            | 98.67%            | 0.0625        |
+| 6    | ResNet101V2      | 100.0%        | 537.5MB            | 98.74%            | 0.0605        |
+
+- ranking based on Trained Model size
+
+### 📢 **Conclusion**
+
+- All models achieve a remarkable test accuracy of 100.0%, demonstrating their effectiveness in classification tasks.
+  - MobileNetV3Small stands out with a compact size of 19.1MB, offering high accuracy while minimizing resource usage, making it suitable for memory-constrained environments.
+  - NASNetMobile and MobileNetV3Large also deliver impressive accuracy with moderate model sizes, providing versatility in deployment scenarios.
+  - InceptionV3, ResNet50V2, and ResNet101V2, although larger in size, exhibit robust performance, with ResNet101V2 achieving the highest training accuracy of 98.74%.
+
+### ✒️ Enhancements done by **Adhithyan VP**
+
+[![LinkedIn](https://img.shields.io/badge/LinkedIn-0077B5?logo=linkedin&logoColor=white)](https://www.linkedin.com/in/adhithyanvp)
+[![X (formerly Twitter) Follow](https://img.shields.io/twitter/follow/AdhiVp3)](https://x.com/AdhiVp3)
 
 **SOCIAL SUMMER OF CODE 2023**
 github link - [aditya0929](https://github.com/aditya0929)

diff --git a/American Sign Language Detection/requirements.txt b/American Sign Language Detection/requirements.txt
@@ -0,0 +1,5 @@
+pandas
+Pillow
+numpy
+tensorflow
+matplotlib