added description

app.py

@@ -92,14 +92,19 @@ Best results are obtained using one of these sentences, which were used during t
 \n\n
 When the binarize option is turned off, model will output propabilities of requested {class} for each patch. When binarize option is turned on
 the model will binarize each propability based on set eval_threshold.
+\n\n
+Each input image is transformed to size 224x224 so it can be processed by ViT. During this transformation, different
+crop_modes and crop_percentage can be selected. The model was trained using crop_mode='center', crop_pct = 0.9. 
+For explanation of different crop_modes, please refer to 
+<a href="https://github.com/huggingface/pytorch-image-models/blob/main/timm/data/transforms_factory.py">this</a> website, lines 155-172.
 """
 demo = gr.Interface(
     query_image, 
     #inputs=[gr.Image(), "text", "checkbox", gr.Slider(0, 1, value=0.25)],
-    inputs=[gr.Image(type='numpy', label='input_img').style(height=200, width=600), "text", "checkbox", gr.Slider(0, 1, value=0.25), 
-            gr.Radio(["center", "squash", "border"], value='center', label='crop_mode'), gr.Slider(0.7, 1, value=1)],  
+    inputs=[gr.Image(type='numpy', label='input_img').style(height=196, width=600), "text", "checkbox", gr.Slider(0, 1, value=0.25), 
+            gr.Radio(["center", "squash", "border"], value='center', label='crop_mode'), gr.Slider(0.7, 1, value=0.9)],  
     #outputs="image",
-    outputs=gr.Image(type='numpy', label='output').style(height=610, width=600),
+    outputs=gr.Image(type='numpy', label='output').style(height=600, width=600),
     title="Object Detection Using Textual Queries",
     description=description,
     examples=[