changed some images

_projects/3_project.md

@@ -2,9 +2,9 @@
 layout: page  
 title: Embedded Toy Detection  
 description: Real-time toy detection for mobile robots using advanced object detection models  
-img:  
+img: assets/img/toy_det_image.png
 importance: 3  
-category: ml_research
+category: ml_research     
 ---
 
 This research project enhances a baseline object detection model through advanced techniques such as pseudo-labelling, data engineering, and knowledge distillation. It develops an efficient and scalable **toy detection system** tailored for robotics applications. Leveraging state-of-the-art models like **Co-DETR**, the system pseudo-labels unannotated data to expand the dataset and train a larger model (**DAMO-YOLO-M**) to improve detection accuracy. The deployment model (**DAMO-YOLO-Tiny**) is distilled from the larger variant for lightweight, real-time inference on resource-constrained robotic platforms.
@@ -22,12 +22,12 @@ This research project enhances a baseline object detection model through advance
 
 <div class="row">
    <div class="col-sm mt-6 mt-md-0">
-       {% include figure.liquid loading="eager" path="assets/img/toy_inference_.png" title="Inference" class="img-fluid rounded z-depth-1" %}
+       {% include figure.liquid loading="eager" path="assets/img/toy_inference.png" title="Inference" class="img-fluid rounded z-depth-1" %}
    </div>
    <div class="col-sm mt-6 mt-md-0"> 
-       {% include figure.liquid loading="eager" path="assets/img/metric_improv_.png" title="Model Improvements" class="img-fluid rounded z-depth-1" %}
+       {% include figure.liquid loading="eager" path="assets/img/metric_improv.png" title="Model Improvements" class="img-fluid rounded z-depth-1" %}
    </div>
-</div>
+</div> 
 <div class="caption">
    Inference and baseline improvements.
 </div>

_projects/5_project.md

@@ -13,6 +13,17 @@ This project explores the use of **Deep Convolutional Generative Adversarial Net
 
 The study successfully generates realistic synthetic images using just 365 training samples, demonstrating the potential of GANs in improving model performance for data-scarce applications. Additionally, a variant of the architecture was applied to generate **anime-style faces**, showcasing the versatility of the approach.  
 
+---
+
+<div class="row">
+   <div class="col-sm mt-6 mt-md-0">
+       {% include figure.liquid path="assets/img/DCGAN_architecture.png" title="Training Dataset Image" class="img-fluid rounded z-depth-1" %}
+   </div>
+</div>
+<div class="caption">
+   Model Architecture of DCGAN
+</div>
+
 --- 
 
 ## Research Highlights