测试集结果文件md5
链接：https://pan.baidu.com/s/1uFOFjj5QdCvaCw-8MYTFhA 提取码：jpmn --来自百度网盘超级会员V5的分享
(base) ➜ visualbert git:(kk) ✗ md5sum logs/vqa_finetune_0831/result.json
22e0a4ce40f0ec92d8fa730b697101cd logs/vqa_finetune_0831/result.json下载提交到竞赛网站上的 Submitted file, 同样检查md5
(base) ➜ visualbert git:(kk) ✗ md5sum logs/vqa_finetune_0831/result.json
22e0a4ce40f0ec92d8fa730b697101cd logs/vqa_finetune_0831/result.json
(base) ➜ visualbert git:(kk) ✗ wget https://evalai.s3.amazonaws.com/media/submission_files/submission_156749/2ae1c2c6-7f72-4dd6-994a-b0adf5867983.json--2021-09-13 16:01:56-- https://evalai.s3.amazonaws.com/media/submission_files/submission_156749/2ae1c2c6-7f72-4dd6-994a-b0adf5867983.json
Resolving evalai.s3.amazonaws.com... 52.217.14.60
Connecting to evalai.s3.amazonaws.com|52.217.14.60|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 20517922 (20M) [application/json]
Saving to: ‘2ae1c2c6-7f72-4dd6-994a-b0adf5867983.json’
2ae1c2c6-7f72-4dd6-994a-b0adf5867983.json 100%[======================================================================================================>] 19.57M 1.70MB/s in 17s
2021-09-13 16:02:30 (1.15 MB/s) - ‘2ae1c2c6-7f72-4dd6-994a-b0adf5867983.json’ saved [20517922/20517922]
(base) ➜ visualbert git:(kk) ✗ md5sum 2ae1c2c6-7f72-4dd6-994a-b0adf5867983.json
22e0a4ce40f0ec92d8fa730b697101cd 2ae1c2c6-7f72-4dd6-994a-b0adf5867983.json
This repository contains code for the following two papers:
-
VisualBERT: A Simple and Performant Baseline for Vision and Language (arxiv) with a short version titiled What Does BERT with Vision Look At? published on ACL 2020.
Under the folder
visualbertis code (the original VisualBERT), where we pre-train a Transformer for vision-and-language (V&L) tasks on image-caption data. -
Unsupervised Vision-and-Language Pre-training Without Parallel Images and Captions published on NAACL 2021.
Under the folder
unsupervised_visualbertis code (Unsupervised VisualBERT), where we pre-train a V&L transformer without aligned image-captions pairs. Rather, we pre-training only using unaligned images and text, and achieve competitive performance with many models supervised with aligned data.
The model VisualBERT has been also integrated into several libararies such as Huggingface Transformer (many thanks to Gunjan Chhablani who made it work) and Facebook MMF.
Thanks~
![dependabot[bot] avatar](https://avatars.githubusercontent.com/in/29110?v=4 "dependabot[bot]")
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent