front page

app.py

@@ -277,41 +277,42 @@ def main():
                  "You can choose between inpainting, segmentation conditioning and re-generating objects, which "
                  "use our custom trained controlnet model."
                  )
-        st.write("### About the dataset")
-        st.write("To make this demo as good as possible, our team spend a lot of time training a custom model. "
-                 "We used the LAION5B dataset to build our custom dataset, which contains 130k images of 15 types of rooms "
-                 "in almost 30 design styles. After fetching all these images, we started adding metadata such as "
-                 "captions (from the BLIP captioning model) and segmentation maps (from the HuggingFace UperNetForSemanticSegmentation model). "
-                 )
-        st.write("### About the model")
-        st.write(
-            "These were then used to train the controlnet model to generate quality interior design images by using "
-            "the segmentation maps and prompts as conditioning information for the model. "
-            "By training on segmentation maps, the enduser has a very finegrained control over which objects they "
-            "want to place in their room. "
-            "The resulting model is then used in a community pipeline that supports image2image and inpainting, "
-            "so the user can keep elements of their room and change specific parts of the image."
-            ""
-        )
-        
-        st.write("### Trivia")
-        st.write("The first time someone uses the demo after startup, the models still need to be loaded into memory. "
-                 "After this initial load, the model is cached as a resource and can be used for all the users. "
-                 "To avoid simultaneous requests, we have implemented a queueing mechanism that ensures that only one "
-                 "user accesses the model at a time (similar to the Gradio framework).\n"
-                 )
-        st.write("To enable the features in the demo, we calculate the underlying segmentation maps and categories that "
-                 "are present in the image. This allows us to hide some of the manual work for the user, and "
-                 "by doing this, the users don't need to make a segmentation map in an external tool. Everything needed can be done within this demo."
-                 )
-        
-        st.write("### News: Fondant - an open source data-centric framework for Foundation model finetuning")
-        st.write("The ML6 team is proud to announce that we are open sourcing our Fondant framework, which is a "
-                 "data-centric framework that allows you to prepare large scale multimodal datasets with ease. We have implemented the components "
-                 "that we used to train this controlnet model in Fondant as an example pipeline, and we are excited to see what you can do with it! In the future we will add a whole library of plug-and-play data preparation components, such as different ML models and filtering steps, in addition to dataset scraping components that connect to LAION5B."
-                 )
-        st.write("The framework is build on top of kubeflow pipelines and abstracts all the complexity of efficient storing and moving of large datasets, so you can focus on implemented just that piece of code that you need without worrying about the rest. We also build it to run on each Cloud provider or VM. You can find the code on our github page: https://github.com/ml6team/fondant.")
-        
+        with st.expander("Useful information"):
+            st.write("### About the dataset")
+            st.write("To make this demo as good as possible, our team spend a lot of time training a custom model. "
+                    "We used the LAION5B dataset to build our custom dataset, which contains 130k images of 15 types of rooms "
+                    "in almost 30 design styles. After fetching all these images, we started adding metadata such as "
+                    "captions (from the BLIP captioning model) and segmentation maps (from the HuggingFace UperNetForSemanticSegmentation model). "
+                    )
+            st.write("### About the model")
+            st.write(
+                "These were then used to train the controlnet model to generate quality interior design images by using "
+                "the segmentation maps and prompts as conditioning information for the model. "
+                "By training on segmentation maps, the enduser has a very finegrained control over which objects they "
+                "want to place in their room. "
+                "The resulting model is then used in a community pipeline that supports image2image and inpainting, "
+                "so the user can keep elements of their room and change specific parts of the image."
+                ""
+            )
+            
+            st.write("### Trivia")
+            st.write("The first time someone uses the demo after startup, the models still need to be loaded into memory. "
+                    "After this initial load, the model is cached as a resource and can be used for all the users. "
+                    "To avoid simultaneous requests, we have implemented a queueing mechanism that ensures that only one "
+                    "user accesses the model at a time (similar to the Gradio framework).\n"
+                    )
+            st.write("To enable the features in the demo, we calculate the underlying segmentation maps and categories that "
+                    "are present in the image. This allows us to hide some of the manual work for the user, and "
+                    "by doing this, the users don't need to make a segmentation map in an external tool. Everything needed can be done within this demo."
+                    )
+            
+            st.write("### News: Fondant - an open source data-centric framework for Foundation model finetuning")
+            st.write("The ML6 team is proud to announce that we are open sourcing our Fondant framework, which is a "
+                    "data-centric framework that allows you to prepare large scale multimodal datasets with ease. We have implemented the components "
+                    "that we used to train this controlnet model in Fondant as an example pipeline, and we are excited to see what you can do with it! In the future we will add a whole library of plug-and-play data preparation components, such as different ML models and filtering steps, in addition to dataset scraping components that connect to LAION5B."
+                    )
+            st.write("The framework is build on top of kubeflow pipelines and abstracts all the complexity of efficient storing and moving of large datasets, so you can focus on implemented just that piece of code that you need without worrying about the rest. We also build it to run on each Cloud provider or VM. You can find the code on our github page: https://github.com/ml6team/fondant.")
+
         st.write("### Testing images")
         st.write("If you don't have any pictures close, you can use one of these images to test the model:")