Open ShiYaya opened 3 years ago
We use all 5,000 images in the Karpathy test split. In my log file, it shows 'using 5000/5000 predictions'.
No test image will be filtered unless the test images' ids are different from those in ground-truth files.
非常感谢你的回复,您可否提供一下您的预测输出,即在 Karpathy test split 上预测的 captions。谢谢
Here I show one example test results on Karpathy test split. It has 5,000 records with image ids.
Hi, thank you for sharing the output example! Is this the output of "Transformer + CIDErBtwReward" model?
Hi, thank you for sharing the output example! Is this the output of "Transformer + CIDErBtwReward" model?
No, it is just an example.
Thank you! Could you share the outputs of "Transformer + CIDErBtwReward" model?
您好,Table 1 中使用 MSCOCO test split, 我想知道,表格中的数据是使用5000 images? 还是经过了过滤,像eval_utils中的注释:“filter results to only those in MSCOCO validation set (will be about a third)”。
您呈现的结果,是过滤之后的结果吗?