Jack Monas
commited on
Commit
·
4b7e1bc
1
Parent(s):
80e3dd0
rules
Browse files
app.py
CHANGED
@@ -1,100 +1,129 @@
|
|
1 |
import streamlit as st
|
2 |
import pandas as pd
|
|
|
3 |
|
4 |
-
|
5 |
-
|
6 |
-
#
|
7 |
-
st.
|
8 |
-
|
9 |
-
|
10 |
-
st.
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
# Overview Section
|
15 |
-
st.markdown("<h1 id='overview'>Overview</h1>", unsafe_allow_html=True)
|
16 |
-
st.header("Welcome")
|
17 |
-
st.write(
|
18 |
-
"Welcome to the World Model Challenge server. This platform hosts three challenges designed to advance research in world models for robotics: "
|
19 |
-
"Compression, Sampling, and Evaluation."
|
20 |
-
)
|
21 |
-
|
22 |
-
# Motivation Section
|
23 |
-
st.markdown("<h1 id='motivation'>Motivation</h1>", unsafe_allow_html=True)
|
24 |
-
st.header("Motivation")
|
25 |
-
st.write(
|
26 |
-
"Real-world robotics faces a constant challenge—environments are dynamic and ever-changing, which makes it difficult to reliably evaluate robot performance. "
|
27 |
-
"World models address this by learning to simulate complex interactions from raw sensor data. These learned simulators allow for robust testing and continuous improvement of robot policies without the limitations of physical testing."
|
28 |
-
)
|
29 |
-
|
30 |
-
# Challenges Section
|
31 |
-
st.markdown("<h1 id='challenges'>Challenges</h1>", unsafe_allow_html=True)
|
32 |
-
st.header("The Challenges")
|
33 |
-
st.subheader("Compression Challenge")
|
34 |
-
st.write(
|
35 |
-
"Train a model to compress our robots' logs effectively while preserving critical details needed to understand and predict future interactions. "
|
36 |
-
"Performance is measured by the model's loss—the lower the loss, the better it captures real-world complexities."
|
37 |
-
)
|
38 |
-
st.subheader("Sampling Challenge")
|
39 |
-
st.write(
|
40 |
-
"Predict a video frame two seconds ahead given a short clip of robot interactions. The goal is to create a coherent and plausible continuation that accurately reflects scene dynamics. "
|
41 |
-
"Submissions are judged by how closely they match the actual frame."
|
42 |
-
)
|
43 |
-
st.subheader("Evaluation Challenge")
|
44 |
-
st.write(
|
45 |
-
"Can you predict a robot's real-world performance without physically deploying it? You'll be given several policies for a specific task and must rank them by expected performance. "
|
46 |
-
"Your ranking is then compared to the true ranking of the policies."
|
47 |
-
)
|
48 |
-
|
49 |
-
# Datasets Section
|
50 |
-
st.markdown("<h1 id='datasets'>Datasets</h1>", unsafe_allow_html=True)
|
51 |
-
st.header("Datasets")
|
52 |
-
st.write(
|
53 |
-
"We offer two key datasets for the 1X World Model Challenge:\n\n"
|
54 |
-
"**Raw Data:** The [world_model_raw_data](https://huggingface.co/datasets/1x-technologies/world_model_raw_data) dataset provides raw sensor data, video logs, and annotated robot state sequences from diverse real-world scenarios. "
|
55 |
-
"It is split into 100 shards—each containing a 512x512 MP4 video, a segment index mapping, and state arrays—and is licensed under CC-BY-NC-SA 4.0.\n\n"
|
56 |
-
"**Tokenized Data:** The [world_model_tokenized_data](https://huggingface.co/datasets/1x-technologies/world_model_tokenized_data) dataset tokenizes raw video sequences using the NVIDIA Cosmos Tokenizer. "
|
57 |
-
"This compact representation is optimal for the Compression Challenge and is released under the Apache 2.0 license."
|
58 |
-
)
|
59 |
-
|
60 |
-
# Scoring Section
|
61 |
-
st.markdown("<h1 id='scoring'>Scoring</h1>", unsafe_allow_html=True)
|
62 |
-
st.header("Scoring")
|
63 |
-
st.write(
|
64 |
-
"Our scoring system rewards performance in all three challenges, with extra emphasis on the Evaluation Challenge. "
|
65 |
-
"A team's final rank is determined by the total points earned across the challenges."
|
66 |
-
)
|
67 |
-
st.subheader("Points Breakdown")
|
68 |
-
col1, col2, col3 = st.columns(3)
|
69 |
-
with col1:
|
70 |
-
st.markdown('<h3 style="margin-left:20px;">Compression</h3>', unsafe_allow_html=True)
|
71 |
-
st.markdown(
|
72 |
-
"- **1st Place:** 10 points\n"
|
73 |
-
"- **2nd Place:** 7 points\n"
|
74 |
-
"- **3rd Place:** 5 points"
|
75 |
)
|
76 |
-
|
77 |
-
|
78 |
-
st.markdown(
|
79 |
-
|
80 |
-
|
81 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
82 |
)
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
|
|
|
|
|
|
|
|
|
|
89 |
)
|
90 |
-
|
|
|
|
|
|
|
91 |
st.write(
|
92 |
-
"
|
93 |
-
"
|
94 |
-
"
|
95 |
-
"
|
96 |
)
|
97 |
-
st.
|
98 |
-
|
99 |
-
"
|
100 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import streamlit as st
|
2 |
import pandas as pd
|
3 |
+
import streamlit.components.v1 as components
|
4 |
|
5 |
+
|
6 |
+
def scoring_section():
|
7 |
+
# Title
|
8 |
+
st.markdown("#### Scoring")
|
9 |
+
|
10 |
+
# Intro text
|
11 |
+
st.write(
|
12 |
+
"Our scoring system rewards performance in all three challenges, with extra emphasis on the Evaluation Challenge. "
|
13 |
+
"A team's final rank is determined by the total points they earn across the challenges."
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
)
|
15 |
+
|
16 |
+
# Points Breakdown in a table
|
17 |
+
st.markdown("#### Points Breakdown")
|
18 |
+
# Create three columns for a more interesting layout
|
19 |
+
col1, col2, col3 = st.columns(3)
|
20 |
+
|
21 |
+
with col1:
|
22 |
+
st.markdown('<h3 style="margin-left:20px;">Compression</h3>', unsafe_allow_html=True)
|
23 |
+
st.markdown(
|
24 |
+
"""
|
25 |
+
- **1st Place**: 10 points
|
26 |
+
- **2nd Place**: 7 points
|
27 |
+
- **3rd Place**: 5 points
|
28 |
+
"""
|
29 |
+
)
|
30 |
+
|
31 |
+
with col2:
|
32 |
+
st.markdown('<h3 style="margin-left:20px;">Sampling</h3>', unsafe_allow_html=True)
|
33 |
+
st.markdown(
|
34 |
+
"""
|
35 |
+
- **1st Place**: 10 points
|
36 |
+
- **2nd Place**: 7 points
|
37 |
+
- **3rd Place**: 5 points
|
38 |
+
"""
|
39 |
+
)
|
40 |
+
|
41 |
+
with col3:
|
42 |
+
st.markdown('<h3 style="margin-left:20px;">Evaluation</h3>', unsafe_allow_html=True)
|
43 |
+
st.markdown(
|
44 |
+
"""
|
45 |
+
- **1st Place**: 20 points
|
46 |
+
- **2nd Place**: 14 points
|
47 |
+
- **3rd Place**: 10 points
|
48 |
+
"""
|
49 |
+
)
|
50 |
+
# Tie-Breakers in an expander for a cleaner layout
|
51 |
+
with st.expander("Tie-Breakers"):
|
52 |
+
st.write(
|
53 |
+
"The overall winner will be the team with the highest total points. "
|
54 |
+
"In the event of a tie, the following tie-breakers will be applied in order:\n\n"
|
55 |
+
"1. Highest Evaluation Challenge score\n"
|
56 |
+
"2. Highest Sampling Challenge score\n"
|
57 |
+
"3. Highest Compression Challenge score\n\n"
|
58 |
+
)
|
59 |
+
|
60 |
+
# Overall Leaderboard Section
|
61 |
+
st.write(
|
62 |
+
"The leaderboard, which shows the total points across all challenges, will go live on **March 10th**. "
|
63 |
+
"Additionally, each challenge—**Compression**, **Sampling**, and **Evaluation**—will have its own leaderboard on their "
|
64 |
+
"respective Hugging Face submission servers."
|
65 |
)
|
66 |
+
|
67 |
+
|
68 |
+
|
69 |
+
def main():
|
70 |
+
st.set_page_config(page_title="World Model Challenge")
|
71 |
+
|
72 |
+
st.title("World Model Challenge")
|
73 |
+
st.markdown("### Welcome")
|
74 |
+
st.write(
|
75 |
+
"Welcome to the World Model Challenge server. This platform hosts three challenges "
|
76 |
+
"designed to advance research in world models for robotics: Compression, Sampling, and Evaluation."
|
77 |
)
|
78 |
+
|
79 |
+
st.markdown("---")
|
80 |
+
|
81 |
+
st.markdown("### Motivation")
|
82 |
st.write(
|
83 |
+
"Real-world robotics faces a fundamental challenge: environments are dynamic and change over time, "
|
84 |
+
"making consistent evaluation of robot performance difficult. World models offer a solution by "
|
85 |
+
"learning to simulate complex real-world interactions from raw sensor data. We believe these learned simulators will enable "
|
86 |
+
"robust evaluation and iterative improvement of robot policies without the constraints of a physical testbed."
|
87 |
)
|
88 |
+
st.markdown("---")
|
89 |
+
|
90 |
+
st.markdown("### The Challenges")
|
91 |
+
|
92 |
+
st.markdown("#### Compression Challenge")
|
93 |
+
st.write(
|
94 |
+
"In the Compression Challenge, your task is to train a model to compress our robots logs effectively while preserving the critical details needed to understand and predict future interactions. Success in this challenge is measured by the loss of your model—the lower the loss, the better your model captures the complexities of real-world robot behavior."
|
95 |
+
)
|
96 |
+
|
97 |
+
st.markdown("#### Sampling Challenge")
|
98 |
+
st.write(
|
99 |
+
"In the Sampling Challenge, your task is to predict a future video frame two seconds in the future given a short clip of robot interactions. The goal is to produce a coherent and plausible continuation of the video, which accurately reflects the dynamics of the scene. Your submission will be judged on how closely it matches the actual frame."
|
100 |
+
)
|
101 |
+
|
102 |
+
st.markdown("#### Evaluation Challenge")
|
103 |
+
st.write(
|
104 |
+
"The Evaluation Challenge tackles the ultimate question: Can you predict a robot's performance in the real world without physically deploying it? In this challenge, you will be provided with many different policies for a specific task. Your task is to rank these policies according to their expected real-world performance. This ranking will be compared with the actual ranking of the policies."
|
105 |
+
)
|
106 |
+
st.markdown("---")
|
107 |
+
|
108 |
+
|
109 |
+
st.markdown("### Datasets")
|
110 |
+
st.write(
|
111 |
+
"We offer two key datasets to support the 1X World Model Challenge:\n\n"
|
112 |
+
"**Raw Data:** The [world_model_raw_data](https://huggingface.co/datasets/1x-technologies/world_model_raw_data) dataset "
|
113 |
+
"provides raw sensor data, video logs, and annotated robot state sequences gathered from diverse real-world scenarios. "
|
114 |
+
"This dataset is split into 100 shards—each containing a 512x512 MP4 video, a segment index mapping, and state arrays—"
|
115 |
+
"and is licensed under CC-BY-NC-SA 4.0.\n\n"
|
116 |
+
"**Tokenized Data:** The [world_model_tokenized_data](https://huggingface.co/datasets/1x-technologies/world_model_tokenized_data) dataset "
|
117 |
+
"tokenizes the raw video sequences generated using the NVIDIA Cosmos Tokenizer. This compact representation of the raw data "
|
118 |
+
"is optimal for the compression challenge and is released under the Apache 2.0 license.\n\n"
|
119 |
+
)
|
120 |
+
|
121 |
+
st.markdown("---")
|
122 |
+
|
123 |
+
|
124 |
+
scoring_section()
|
125 |
+
|
126 |
+
|
127 |
+
|
128 |
+
if __name__ == '__main__':
|
129 |
+
main()
|