cleaned up readme slightly

app.py

@@ -664,9 +664,6 @@ def track(
 
     video_input = torch.tensor(video_input).unsqueeze(0).to(dtype)
     print("1 torch.cuda.memory_allocated: %.1fGB"%(torch.cuda.memory_allocated(0)/1024/1024/1024))
-
-    # model = torch.hub.load("facebookresearch/co-tracker", "cotracker3_online")
-    # model = model.to(device)
     
     from nets.alltracker import Net
     model = Net(16)
@@ -878,9 +875,9 @@ with gr.Blocks() as demo:
     gr.Markdown("# ⚡ AllTracker: Efficient Dense Point Tracking at High Resolution")
     gr.Markdown("<div style='text-align: left;'> \
     <p>Welcome to <a href='https://alltracker.github.io/' target='_blank'>AllTracker</a>! This demo runs our model to perform all-pixel tracking in a video of your choice.</p> \
-    <p>To get started, simply upload your <b>.mp4</b> video, or click on one of the example videos. The shorter the video, the faster the processing. We recommend submitting videos under 20 seconds long.</p> \
-    <p>After picking a video, click \"Submit\" to load the frames into the app, and optionally choose a frame (using the slider), and then click \"Track\".</p> \
-    <p>For full info on how this works, check out our <a href='https://github.com/aharley/alltracker/' target='_blank'>GitHub Repo</a>!</p> \
+    <p>To get started, simply upload your <b>.mp4</b> video, or select one of the example videos. The shorter the video, the faster the processing. We recommend submitting videos under 20 seconds long.</p> \
+    <p>After picking a video, click \"Submit\" to load the frames into the app, and optionally choose a query frame (using the slider), and then click \"Track\".</p> \
+    <p>For full info on how this works, check out our <a href='https://github.com/aharley/alltracker/' target='_blank'>GitHub repo</a>, or <a href='https://arxiv.org/abs/2506.07310' target='_blank'>paper</a>.</p> \
     <p>Initial code for this Gradio app came from LocoTrack and CoTracker -- big thanks to those authors!</p> \
     </div>"
     )