"""Layout primitives specific to HF Space presentation.""" from __future__ import annotations import streamlit as st ACCENT_ORANGE = "#FF6B35" ACCENT_CYAN = "#00B8D9" def apply_space_theme() -> None: st.set_page_config( page_title="WarehouseEye", page_icon="📦", layout="wide", initial_sidebar_state="expanded", ) st.markdown( f""" """, unsafe_allow_html=True, ) def render_hero(github_url: str) -> None: st.markdown( """

WarehouseEye

Operational intelligence for warehouse CCTV — on AMD MI300X

Built on AMD MI300X Powered by Qwen3-VL MIT License

""", unsafe_allow_html=True, ) st.link_button("View on GitHub", github_url, use_container_width=True) def render_how_it_works() -> None: with st.expander("How it works", expanded=False): st.markdown( """

The problem: warehouse operators often review long CCTV footage manually, which is slow, expensive, and easy to miss critical safety or process events.

The solution: WarehouseEye is an open-source pipeline that combines tracking with a vision-language model to convert long videos into searchable operational timelines.

Why AMD MI300X: its 192 GB HBM memory enables high-capacity multimodal inference on a single node, avoiding multi-GPU coordination complexity for this workload.

""", unsafe_allow_html=True, ) def render_space_footer(github_url: str, paper_url: str, twitter_url: str) -> None: links: list[str] = [ f'GitHub repo', f'Author X/Twitter', ] if paper_url: links.insert( 1, f'Technical paper', ) st.markdown( f""" """, unsafe_allow_html=True, )