EMCompress: A Benchmark for Endomorphic Multimodal Compression on Long Cooking Videos

EMCompress is the first benchmark for evaluating Endomorphic Multimodal Compression (EMC). The dataset was released by LordUky in May 2026 and includes reproduction code. The associated paper was accepted to ACL 2026 Findings.

Use Cases

• Benchmarking multimodal compression models based on the described endomorphic transformation task.
• Evaluating video-question pair compression performance on long-form content.
• Training models for efficient representation of cooking videos and associated textual queries.

Strengths

• Serves as the first dedicated benchmark for the Endomorphic Multimodal Compression task.
• Associated with a peer-reviewed paper accepted to ACL 2026 Findings.
• Reproduction code is available alongside the dataset.

Limitations

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

Provenance

Source

LordUky on HuggingFace

Freshness

Last updated 2026-05-12 01:12:58; freshness should be verified.