InternVL icon indicating copy to clipboard operation
InternVL copied to clipboard

Refer Expression Comprehension RefCOCO 验证结果不一致

Open yanzaaaasa opened this issue 1 year ago • 2 comments

GPUS=8 sh evaluate.sh refcoco 测试脚本 InternVL 1.5 测试出的指标如下,请问一下是代码有问题吗 ['model/InternVL-Chat-V1-5', 'refcoco_val', 'Precision @ 1: 0.9019752630607347 \n'] ['model/InternVL-Chat-V1-5', 'refcoco_testA', 'Precision @ 1: 0.9284072830121973 \n'] ['model/InternVL-Chat-V1-5', 'refcoco_testB', 'Precision @ 1: 0.8563297350343474 \n'] ['model/InternVL-Chat-V1-5', 'refcoco+_val', 'Precision @ 1: 0.8406767057073805 \n'] ['model/InternVL-Chat-V1-5', 'refcoco+_testA', 'Precision @ 1: 0.8899755501222494 \n'] ['model/InternVL-Chat-V1-5', 'refcoco+_testB', 'Precision @ 1: 0.785232153814686 \n'] ['model/InternVL-Chat-V1-5', 'refcocog_val', 'Precision @ 1: 0.8654003267973857 \n'] ['model/InternVL-Chat-V1-5', 'refcocog_test', 'Precision @ 1: 0.8680483232659862 \n']

释放出来的指标如下。 InternVL−Chat−V1.5 26B 91.4 93.7 87.1 87.0 92.3 80.9 88.5 89.3

yanzaaaasa avatar May 23 '24 04:05 yanzaaaasa

是不是没有加 --dynamic

GPUS=8 sh evaluate.sh XXXX refcoco --dynamic

czczup avatar May 23 '24 07:05 czczup

是的,加上 --dynamic,测出来的结果一致了,感谢~

yanzaaaasa avatar May 23 '24 11:05 yanzaaaasa