search

rizavelioglu / hateful_memes-hate_detectron

Detecting Hate Speech in Memes Using Multimodal Deep Learning Approaches: Prize-winning solution to Hateful Memes Challenge. https://arxiv.org/abs/2012.12975
https://rizavelioglu.github.io/publication/2021-04-msc-thesis
MIT License
54 stars 19 forks source link

why we could use mmf_run to run the project? #4

Open fengmingfeng opened 2 years ago

fengmingfeng commented 2 years ago

why we could use mmf_run to run the project? what is th main function of the project?

rizavelioglu commented 2 years ago

Hi Feng, I am sorry, I am unable to understand the issue. could you please provide more information

fengmingfeng commented 2 years ago

Hi rizavelioglu,     thank you for you reply.when i  reproduce your code of (https://github.com/yatingtian/hateful-meme/blob/main/visualbert/Detectron.ipynb),and i successfully extract image feature,then i train visualbert model but i occur follow problem:     feats_dir = "/content/gdrive/MyDrive/feature" train_dir="/content/gdrive/MyDrive/Colab Notebooks/hateful_memes/defaults/annotations/train.jsonl" dev_unseen="/content/gdrive/MyDrive/Colab Notebooks/hateful_memes/defaults/annotations/dev_unseen.jsonl" test_unseen="/content/gdrive/MyDrive/Colab Notebooks/hateful_memes/defaults/annotations/test_unseen.jsonl"

    !mmf_run config="projects/visual_bert/configs/hateful_memes/from_coco.yaml" \         model="visual_bert" \         dataset=hateful_memes \         run_type=train_val \         checkpoint.max_to_keep=1 \         checkpoint.resume_zoo=visual_bert.pretrained.cc.small_fifty_pc \         training.tensorboard=True \         training.checkpoint_interval=50 \         training.evaluation_interval=50 \         training.max_updates=3000 \         training.log_interval=100 \         dataset_config.hateful_memes.max_features=120 \         dataset_config.hateful_memes.annotations.train[0]=$train_dir \         dataset_config.hateful_memes.annotations.val[0]=$dev_unseen \         dataset_config.hateful_memes.annotations.test[0]=$test_unseen \         dataset_config.hateful_memes.features.train[0]=$feats_dir \         dataset_config.hateful_memes.features.val[0]=$feats_dir \         dataset_config.hateful_memes.features.test[0]=$feats_dir \         training.lr_ratio=0.3 \         training.use_warmup=True \         training.batch_size=32 \         optimizer.params.lr=5.0e-05 \         env.save_dir=./sub1 \         env.tensorboard_logdir=logs/fit/sub1 \

  this is the part of image feature

i don't konw  whether !mmf_run is run mmf_cli/run.py,and how to solve the previous problem? thank you for your reply. ------------------ 原始邮件 ------------------ 发件人: "rizavelioglu/hateful_memes-hate_detectron" @.>; 发送时间: 2022年7月19日(星期二) 下午3:39 @.>; @.**@.>; 主题: Re: [rizavelioglu/hateful_memes-hate_detectron] why we could use mmf_run to run the project? (Issue #4)

Hi Feng, I am sorry, I am unable to understand the issue. could you please provide more information

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

fengmingfeng commented 2 years ago

Hi rizavelioglu,     when i reproduce your code.I occur some problem,as follows are the problem? command: !mmf_run config="/kaggle/working/mmf/projects/vilbert/configs/hateful_memes/from_cc.yaml" \         model="vilbert" \         dataset=hateful_memes \         run_type=train_val \         checkpoint.max_to_keep=1 \         checkpoint.resume_zoo=vilbert.pretrained.cc.full \         training.tensorboard=True \         training.checkpoint_interval=50 \         training.evaluation_interval=50 \         training.max_updates=3000 \         training.log_interval=100 \         dataset_config.hateful_memes.max_features=120 \         dataset_config.hateful_memes.annotations.train[0]=$train_dir \         dataset_config.hateful_memes.annotations.val[0]=$dev_unseen \         dataset_config.hateful_memes.annotations.test[0]=$test_unseen \         dataset_config.hateful_memes.features.train[0]=$feats_dir \         dataset_config.hateful_memes.features.val[0]=$feats_dir \         dataset_config.hateful_memes.features.test[0]=$feats_dir \         training.lr_ratio=0.3 \         training.use_warmup=True \         training.batch_size=32 \         optimizer.params.lr=5.0e-05 \         env.save_dir=./sub4 \         env.tensorboard_logdir=logs/fit/sub4

but exist some problem?

could you please give me some help?

------------------ 原始邮件 ------------------ 发件人: "rizavelioglu/hateful_memes-hate_detectron" @.>; 发送时间: 2022年7月19日(星期二) 下午3:39 @.>; @.**@.>; 主题: Re: [rizavelioglu/hateful_memes-hate_detectron] why we could use mmf_run to run the project? (Issue #4)

Hi Feng, I am sorry, I am unable to understand the issue. could you please provide more information

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the threaMessage ID: @.***>