Merge branch 'main' of https://huggingface.co/spaces/krishnapal2308/eye_for_blind

app.py

@@ -11,13 +11,13 @@ warnings.filterwarnings('ignore')
 # Define problem statement
 problem_statement = """
 ### Problem Statement
-Visually impaired individuals face challenges in understanding image content. This project aims to address this issue by generating descriptive spoken captions for images, leveraging CNNs and RNNs for feature extraction and sequence generation, respectively. The model is trained on the Flickr8K dataset and extended with an attention mechanism for enhanced accessibility.
+This project aims to develop a deep learning model to verbally describe image contents for the visually impaired using caption generation with an attention mechanism on the Flickr8K dataset. Inspired by the "Show, Attend and Tell: Neural Image Caption Generation with Visual Attention" paper, the model utilizes a CNN-RNN architecture to extract image features and generate captions, facilitating accessibility. The Kaggle dataset comprises 8,000 images, each paired with five descriptive captions, enabling comprehensive understanding of image content.
 """
 
 # Define solution overview
 solution_overview = """
 ### Solution Overview
-The basic model, trained for a limited duration without extensive hyperparameter tuning, primarily focuses on exploring subclassing techniques. To improve inference quality, Vit-GPT2 architecture is integrated. [Visit the Kaggle notebook](https://www.kaggle.com/code/krishna2308/eye-for-blind) for implementation details.
+The basic model, **trained for only 20 epochs without extensive hyperparameter tuning,** primarily focuses on exploring the integration of the attention mechanism with the Encoder-Decoder architecture for image processing utilizing subclassing. To improve inference quality, Vit-GPT2 architecture is integrated. [Visit the Kaggle notebook](https://www.kaggle.com/code/krishna2308/eye-for-blind) for implementation details.
 """
 
 # Define real-life scenario application
@@ -30,7 +30,7 @@ While this current implementation may not support real-time processing, the pote
 def process_image_and_generate_output(image, model_selection):
     if image is None:
         return "Please select an image", None
-    if model_selection == 'Basic Model':
+    if model_selection == "Basic Model (Results won't be good)":
         result = inference_script.evaluate(image)
         pred_caption = ' '.join(result).rsplit(' ', 1)[0]
         pred_caption = pred_caption.replace('<unk>', '')
@@ -57,16 +57,16 @@ def process_image_and_generate_output(image, model_selection):
 
 sample_images = [
     [os.path.join(os.path.dirname(__file__), "sample_images/1.jpg"), "ViT-GPT2"],
-    [os.path.join(os.path.dirname(__file__), "sample_images/1.jpg"), 'Basic Model'],
+    [os.path.join(os.path.dirname(__file__), "sample_images/1.jpg"), "Basic Model (Results won't be good)"],
     [os.path.join(os.path.dirname(__file__), "sample_images/3.jpg"), "ViT-GPT2"],
-    [os.path.join(os.path.dirname(__file__), "sample_images/3.jpg"), 'Basic Model']
+    [os.path.join(os.path.dirname(__file__), "sample_images/3.jpg"), "Basic Model (Results won't be good)"]
 ]
 
 # Create a dropdown to select sample image
 image_input = gr.Image(label="Upload Image")
 
 # Create a dropdown to choose the model
-model_selection_input = gr.Radio(["Basic Model",
+model_selection_input = gr.Radio(["Basic Model (Results won't be good)",
                                   "ViT-GPT2"],
                                  label="Choose Model")