import gradio as gr
import spaces
from src.utils import *
from src.backend import *
theme = gr.themes.Soft(primary_hue="cyan", secondary_hue="zinc", font=[gr.themes.GoogleFont("Source Sans 3", weights=(400, 600)),'arial'])
with gr.Blocks(theme=theme) as demo:
with gr.Column(elem_classes="header"):
gr.HTML('
')
gr.Markdown("# 🔵 COP-GEN-Beta: Unified Generative Modelling of COPernicus Imagery Thumbnails")
gr.Markdown("### Miguel Espinosa, Valerio Marsocci, Yuru Jia, Elliot J. Crowley, Mikolaj Czerkawski")
gr.HTML('
')
#gr.Markdown('[[Website](https://miquel-espinosa.github.io/cop-gen-beta/)] [GitHub](https://github.com/miquel-espinosa/COP-GEN-Beta)] [[Model](https://huggingface.co/mespinosami/COP-GEN-Beta) [[Dataset](https://huggingface.co/Major-TOM)]')
gr.Markdown('> ## ⚠️ NOTE: This is a prototype Beta model of COP-GEN. It is based on image thumbnails of [Major TOM](https://huggingface.co/Major-TOM) and does not yet support raw source data. The hillshade visualisation is used for elevation. The full model COP-GEN is coming soon.')
with gr.Column(elem_classes="Main app"):
with gr.Accordion("Instructions", open=False) as abstract:
gr.Markdown("1. **Generate**: Click the `🏭 Generate` button to synthesize the output **without any conditions**. The outputs will be shown below - and that's it, you've generated your first sample! 🧑🎨️"+
"
2. **Optionally, define input**: If you want to condition your generation, you can upload your thumbnails manually or you can `🔄 Load` a random sample from Major TOM by clicking the button."+
"
3. **Select conditions**: Each input image can be used as a **conditioning** when it's loaded into the inputs panel. The modalities you wish to generate should have no content in the input panel (you can empty each element by clicking `x` in the top right corner of the image)."+
"
4. **Additional Options**: You can control the number of generation steps (higher number might produce better quality, but will take more time), or set a fixed seed (for reproducible results)."+
"
5. You can also reuse any of the generated samples as input to the model by clicking `♻️ Reuse`")
with gr.Column():
generate_button = gr.Button("🏭 Generate", variant="primary")
gr.Markdown("## Outputs")
with gr.Row():
s2l1c_output = gr.Image(label="S2 L1C (Optical - Top of Atmosphere)", interactive=False)
s2l2a_output = gr.Image(label="S2 L2A (Optical - Bottom of Atmosphere)", interactive=False)
s1rtc_output = gr.Image(label="S1 RTC (SAR)", interactive=False)
dem_output = gr.Image(label="DEM (Elevation)", interactive=False)
with gr.Row():
s2l1c_reuse_button = gr.Button("♻️ Reuse S2 L1C as input ⤵️", variant="primary")
s2l2a_reuse_button = gr.Button("♻️ Reuse S2 L2A as input ⤵️", variant="primary")
s1rtc_reuse_button = gr.Button("♻️ Reuse S1 RTC as input ⤵️", variant="primary")
dem_reuse_button = gr.Button("♻️ Reuse DEM as input ⤵️", variant="primary")
gr.Markdown("---")
with gr.Row():
gr.Markdown("## (Optional) Input Conditions")
load_button = gr.Button("🔄 Load a random sample from Major TOM 🗺", variant="secondary")
with gr.Row():
s2l1c_input = gr.Image(label="S2 L1C (Optical - Top of Atmosphere)", interactive=True)
s2l2a_input = gr.Image(label="S2 L2A (Optical - Bottom of Atmosphere)", interactive=True)
s1rtc_input = gr.Image(label="S1 RTC (SAR)", interactive=True)
dem_input = gr.Image(label="DEM (Elevation)", interactive=True)
gr.Markdown('### Ready? Go back up and press `🏭 Generate` again!')
with gr.Accordion("Advanced Options", open=False) as advanced_options:
num_inference_steps_slider = gr.Slider(minimum=10, maximum=1000, step=10, value=10, label="Inference Steps")
with gr.Row():
seed_number = gr.Number(value=6378, label="Seed")
seed_checkbox = gr.Checkbox(value=True, label="Random")
with gr.Accordion("Abstract", open=False) as abstract:
gr.HTML('
')
gr.Markdown("In remote sensing, multi-modal data from various sensors capturing the same scene offers rich opportunities, but learning a unified representation across these modalities remains a significant challenge. Traditional methods have often been limited to single or dual-modality approaches. In this paper, we introduce COP-GEN-Beta, a generative diffusion model trained on optical, radar, and elevation data from the Major TOM dataset. What sets COP-GEN-Beta apart is its ability to map any subset of modalities to any other, enabling zero-shot modality translation after training. This is achieved through a sequence-based diffusion transformer, where each modality is controlled by its own timestep embedding. We extensively evaluate COP-GEN-Beta on thumbnail images from the Major TOM dataset, demonstrating its effectiveness in generating high-quality samples. Qualitative and quantitative evaluations validate the model's performance, highlighting its potential as a powerful pre-trained model for future remote sensing tasks.")
load_button.click(
fn=sample_shuffle,
outputs=[s2l1c_input,s2l2a_input,s1rtc_input,dem_input]
)
generate_button.click(
fn=generate_output,
inputs=[s2l1c_input,
s2l2a_input,
s1rtc_input,
dem_input,
num_inference_steps_slider, seed_number, seed_checkbox],
outputs=[s2l1c_output, s2l2a_output, s1rtc_output, dem_output],
)
def pass_value(value):
return value
s2l1c_reuse_button.click(fn=pass_value, inputs=[s2l1c_output],outputs=[s2l1c_input])
s2l2a_reuse_button.click(fn=pass_value, inputs=[s2l2a_output],outputs=[s2l2a_input])
s1rtc_reuse_button.click(fn=pass_value, inputs=[s1rtc_output],outputs=[s1rtc_input])
dem_reuse_button.click(fn=pass_value, inputs=[dem_output],outputs=[dem_input])
demo.queue().launch(share=True)