X-Stream: Multi-Stream Video Question Answering Benchmark

X-Stream is a multi-stream video question answering benchmark for streaming understanding. Each sample contains a merged multi-stream video. The dataset was created by researchers from MMLab at The Chinese University of Hong Kong and Huawei Inc., and was last updated on HuggingFace in May 2026.

Use Cases

• Benchmarking model performance on multi-stream video question answering based on the described task.
• Training models for streaming video understanding based on the benchmark's focus.
• Researching the capabilities of MLLMs as multiplexers for multi-modal streams based on the paper's theme.

Strengths

• Designed specifically for multi-stream video understanding, a complex task.
• Created by a collaboration between an academic lab (MMLab, CUHK) and an industry partner (Huawei).

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file formats are unknown, which may limit suitability assessment.

Provenance

Source

MMLab, The Chinese University of Hong Kong; Huawei Inc.

Freshness

Last updated 2026-05-28 06:53:42; freshness should be verified.