Existing machine translation and multimodal translation benchmarks mainly focus on text-only inputs or short examples with tightly aligned image-text pairs. Academic lectures present a different challenge: spoken explanations, presentation slides, and domain-specific terminology interact across long-form discourse. LecTrans is constructed from 383 hours of expert-taught online lectures and provides transcripts, presentation slides, and an eight-language parallel evaluation subset covering English and seven target languages. The 162-minute translation subset spans 17 subjects and contains human-corrected English transcripts, verified slide text, and human-verified translations of both transcripts and slides. We define lecture translation as two complementary tasks: transcript translation for spoken lecture content and slide translation for visually presented text. We report text-only and multimodal baselines, showing that noisy transcripts, loosely aligned slide context, and domain-specific content remain challenging for current systems. LecTrans provides a testbed for evaluating multimodal translation in long-form academic lectures.