Add VDEBench evaluation results

#28
by hongzhuyi - opened

Hi πŸ‘‹ this PR adds this model's VDEBench scores to .eval_results/, so they show up on the model page and the VDEBench leaderboard.

VDEBench evaluates instruction-guided editing of dense English/Chinese document pages (3,506 examples). Each score is the equally weighted mean of IoU, CDM, BLEU-4 and TEDS (0–1, higher is better):

Task Score ↑
local (edited region, headline) 0.706
global (full page) 0.786

These results come from the VDEBench team's new full evaluation of BAGEL-7B-MoT. Detailed results.

BAGEL was evaluated in ordinary image-editing mode, with thinking disabled.

Thanks!

Ready to merge
This branch is ready to get merged automatically.

Sign up or log in to comment