README.md

metadata

title: Demo
emoji: 🔥
colorFrom: purple
colorTo: yellow
sdk: gradio
sdk_version: 3.0.15
app_file: app.py
pinned: false

Visual Interpretation of a Hybrid Model

Building a hybrid model with EfficientNet and Swin Transformer, we have tried to inspect the visual interpretations of a CNN and Transformer blocks of a hybrid model (CNN + Swin Transformer) with the GradCAM technique. As a result, it appears that the transformer blocks are capable of globally refining feature activation across the relevant object, as opposed to the CNN, which is more focused on operating locally. However, the approach that will be shown here, is experimental. The workflow probably can generate a more meaningful modeling approach. The model is trained on tf_flowers dataset, a multi-class classification problem.