ChartBench evaluation
Closed this issue · 5 comments
Hello! I wonder what metrics you have used in Table 3 for ChartBench: accp, npa, or the weighted sum of them?
Hello! I used your code to try ChartMoE. There is a bug about the vision encoder as follows:
Input image size (490 * 490) doesn't match model (336 * 336).
Could you please help to check your model and configuration? Thank you!
Hi, @zli999. Sorry, I have not encountered this problem. Please download the weights and configs on this huggingface link: https://huggingface.co/IDEA-FinAI/chartmoe. Then change your ChartMOE_HF_PATH as shown in the README of this repo.
If you can give me more details to reproduce your bug, that would be really better~. Thanks for you attention! If any question exists (like the version of some python package?), I'm very willing to communicate with you～!
As for the ChartBench result, both Acc+ and NQA are computed~. We will sort out the chartbench evaluation code and open source it later. Thanks for your attention!
It works now! Thx! A great work!
Hi, @zli999 . #6 also encountered this problem. Can you share some details to solve it with @Lv996331209 and me? I will also add this bug into FAQs. Thanks for your help~