gif

app.py

@@ -77,7 +77,9 @@ def main():
         "Welcome to the World Model Challenge. This platform hosts three challenges "
         "designed to advance research in world models for robotics: Compression, Sampling, and Evaluation."
     )
-
+    st.write(
+        "This challenge is part of the Autonomous Grand Challenge 2025, held in conjunction with the CVPR 2025 Workshop (confirmed) and ICCV 2025 Workshop (tentative). With a total cash prize pool of up to USD 100,000, individual track awards exceeding USD 15,000, and travel grants available, we invite you to push the boundaries of robotics research and simulation."
+    )
 
     st.markdown("---")
 
@@ -175,13 +177,13 @@ def main():
         """
     **General Guidelines:**
     - The use of publicly available datasets and pretrained weights is allowed. The use of private datasets or pretrained weights is prohibited.
-    - You may use future actions to condition future frame predictions.
+    - You may use future actions to guide your frame predictions, but you must not use any actual future frames during inference.
     - There is no limit on the inference time for any of the challenges.
     - Naive nearest-neighbor retrieval combined with seeking ahead to the next frames from the training set may yield reasonable performance but is not permitted in solutions.
 
     **Submissions:**
-    - All submissions must be reproducible. Include code, configuration files, and any necessary instructions to replicate your results.
-    - The leaderboard will display results on a public test set; however, the final winner will be determined based on performance on a private test set.
+    - All submissions must be reproducible. Please expect to include code, configuration files, and any necessary instructions to replicate your results.
+    - The leaderboard will display results on a public test set. However, the final winner will be determined based on performance on a private test set.
 
     **Eligibility:**
     - Prizes cannot be awarded to individuals in U.S. sanctioned countries. We reserve the right to withhold prizes if a submission violates the spirit of the challenge.
@@ -195,8 +197,17 @@ def main():
 
     st.markdown("## Already Started Working on These Challenges?")
     st.write(
-        "If you've already begun work on any of the challenges, we encourage you to share your progress with the community. "
-        "Connect with other participants via our Discord channel or GitHub repository to exchange ideas, get feedback, and collaborate."
+            """
+                Before partnering with Open Drive Lab to launch the World Model Challenge at CVPR 2025 and (tentatively) ICCV 2025, we had already released the Compression and Sampling challenges publicly, offering $10K for each. However, due to the rapid progress in the field, we have updated our competitions.
+
+                For the Compression Challenge, we initially used the MAGVIT spatial tokenizer to compare participant losses. We are now switching to the highly performant Cosmos spatial-temporal tokenizer as our new standard.
+
+                For the Sampling Challenge, the requirement has changed from predicting a frame 0.5 seconds in the future to predicting a frame two seconds ahead, in order to encourage more robust solutions.
+
+                Additionally, a previous rule prohibited the use of future actions to condition frame predictions; this restriction has been lifted, and using future actions is now allowed.
+
+                Please note that we will continue to honor solutions for the Compression Challenge $10K award submitted using the MAGVIT tokenizer—if you achieve a loss below 8.0 on our held-out test set—for six months from March 1, 2025. However, these solutions will not be eligible for the CVPR or ICCV competitions.
+            """
     )
 
     st.markdown("---")