[Question] Clarification on video preprocessing for evaluation — HEVC selector, unsupported codecs, and MV+residual vs bitcost

Hi, I'm trying to reproduce the results in Table 4 and have a few questions about the evaluation pipeline:

1. HEVC selector in Table 4: Does "HEVC selector" mean all videos are first re-encoded to HEVC, and then the codec primitives (MV + residual) are extracted from the HEVC bitstream? Or does it mean the selector only works on videos already in HEVC format, with other formats falling back to frames mode?

2. Unsupported codecs in benchmarks: Some standard benchmarks contain videos in formats not natively supported by HEVC/H.264 pipelines (e.g., VP6F in NExTQA, MPEG-4 in MVBench). How were these handled during evaluation? Were they transcoded to HEVC, or evaluated with a different backend?

3. MV+residual vs bitcost: The paper describes the importance map as "MV magnitude + residual energy", but the codebase uses bitcost. Which method was used for the Table 4 numbers? If bitcost was used, could you clarify the relationship between the two?

Thanks!

Start by tracing the evaluation pipeline used for Table 4 and the code path that selects HEVC, frames, MV+residual, or bitcost. No file, test, or entry point is named in the issue; completion requires clarification of transcoding and unsupported-codec handling, plus which importance-map method produced the reported numbers.