first commit

.gitattributes

@@ -53,3 +53,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.jpg filter=lfs diff=lfs merge=lfs -text
 *.jpeg filter=lfs diff=lfs merge=lfs -text
 *.webp filter=lfs diff=lfs merge=lfs -text
+*.json filter=lfs diff=lfs merge=lfs -text
+*.jsonl filter=lfs diff=lfs merge=lfs -text
+*.csv filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,115 @@
+# Large Language Models are Temporal and Causal Reasoners for Video Question Answering 
+
+This is the official implementation of Flipped-VQA (EMNLP 2023) ([arxiv](https://arxiv.org/abs/2310.15747)) ([demo](https://ikodoh.github.io/flipped_vqa_demo.html)).
+
+> Dohwan Ko<sup>1*</sup>, Ji Soo Lee<sup>1*</sup>, Wooyoung Kang<sup>2</sup>, Byungseok Roh<sup>2</sup>, Hyunwoo J. Kim<sup>1</sup>.
+>
+><sup>1</sup>Department of Computer Science and Engineering, Korea University   <sup>2</sup>Kakao Brain
+
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/large-language-models-are-temporal-and-causal/video-question-answering-on-next-qa)](https://paperswithcode.com/sota/video-question-answering-on-next-qa?p=large-language-models-are-temporal-and-causal) [![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/large-language-models-are-temporal-and-causal/video-question-answering-on-situated)](https://paperswithcode.com/sota/video-question-answering-on-situated?p=large-language-models-are-temporal-and-causal) [![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/large-language-models-are-temporal-and-causal/video-question-answering-on-dramaqa)](https://paperswithcode.com/sota/video-question-answering-on-dramaqa?p=large-language-models-are-temporal-and-causal) [![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/large-language-models-are-temporal-and-causal/video-question-answering-on-vlep)](https://paperswithcode.com/sota/video-question-answering-on-vlep?p=large-language-models-are-temporal-and-causal) [![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/large-language-models-are-temporal-and-causal/video-question-answering-on-tvqa)](https://paperswithcode.com/sota/video-question-answering-on-tvqa?p=large-language-models-are-temporal-and-causal)
+
+<div align="center">
+  <img src="asset/main.png" width="900px" />
+</div>
+
+## Setup
+To install requirements, run:
+```
+git clone https://github.com/mlvlab/Flipped-VQA.git
+cd Flipped-VQA
+mkdir pretrained
+conda create -n flipped-vqa python=3.8
+conda activate flipped-vqa
+sh setup.sh
+```
+
+## Dataset & LLaMA Preparation
+
+* You can download our preprocessed datasets (NExT-QA, STAR, DramaQA, VLEP and TVQA) in [huggingface](https://huggingface.co/datasets/ikodoh/Flipped-VQA-Data) (We also provide the fine-tuned model on each dataset).
+
+```
+git lfs install
+git clone https://huggingface.co/datasets/ikodoh/Flipped-VQA-Data
+mv ./Flipped-VQA-Data/data ./
+mv ./Flipped-VQA-Data/checkpoint ./
+unzip ./data/tvqa/tvqa_subtitles.zip -d ./data/tvqa
+rm -rf Flipped-VQA-Data ./data/tvqa/tvqa_subtitles.zip
+```
+
+* You can download original LLaMA at [here](https://github.com/facebookresearch/llama/tree/llama_v1), and put the checkpoint in ```./pretrained```.
+
+```
+./pretrained
+   └─ llama
+       |─ 7B
+       |   |─ consolidated.00.pth
+       |   └─ params.json
+       |─ 13B
+       |   :
+       |─ 33B
+       |   :
+       └─ tokenizer.model
+```
+
+## Training LLaMA-VQA (LLaMA + Flipped-VQA)
+
+### NExT-QA
+
+```
+torchrun --rdzv_endpoint 127.0.0.1:1234 --nproc_per_node 4 train.py --model 7B \
+--max_seq_len 128 --batch_size 8 --epochs 5 --warmup_epochs 2 --bias 3.5 --tau 100. --max_feats 10 --dataset nextqa \
+--blr 9e-2 --weight_decay 0.14 --output_dir ./checkpoint/nextqa --accum_iter 2 --vaq --qav
+```
+
+### STAR
+
+```
+torchrun --rdzv_endpoint 127.0.0.1:1234 --nproc_per_node 4 train.py --model 7B \
+--max_seq_len 128 --batch_size 8 --epochs 5 --warmup_epochs 2 --bias 3 --tau 100. --max_feats 10 --dataset star \
+--blr 9e-2 --weight_decay 0.16 --output_dir ./checkpoint/star --accum_iter 1 --vaq --qav
+```
+
+### DramaQA
+
+```
+torchrun --rdzv_endpoint 127.0.0.1:1234 --nproc_per_node 4 train.py --model 7B \
+--max_seq_len 384 --batch_size 2 --epochs 5 --warmup_epochs 2 --bias 3 --tau 100. --max_feats 10 --dataset dramaqa \
+--blr 9e-2 --weight_decay 0.10 --output_dir ./checkpoint/dramaqa --accum_iter 8 --vaq --qav
+```
+
+### VLEP
+
+```
+torchrun --rdzv_endpoint 127.0.0.1:1234 --nproc_per_node 4 train.py --model 7B \
+--max_seq_len 256 --batch_size 4 --epochs 5 --warmup_epochs 2 --bias 3 --tau 100. --max_feats 10 --dataset vlep \
+--blr 6e-2 --weight_decay 0.20 --output_dir ./checkpoint/vlep --accum_iter 8 --sub --qav
+```
+
+### TVQA
+
+```
+torchrun --rdzv_endpoint 127.0.0.1:1234 --nproc_per_node 8 train.py --model 7B \
+--max_seq_len 650 --batch_size 1 --epochs 5 --warmup_epochs 2 --bias 3 --tau 100. --max_feats 10 --dataset tvqa \
+--blr 7e-2 --weight_decay 0.02 --output_dir ./checkpoint/tvqa --dataset tvqa --accum_iter 4 --sub --vaq --qav
+```
+
+The fine-tuned checkpoints on each dataset are [here](https://huggingface.co/datasets/ikodoh/Flipped-VQA-Data).
+
+## Evaluation
+From the training command, simply replace ```train.py``` with ```eval.py``` and add ```--resume ./your/checkpoint.pth```.
+
+## Acknowledgements
+
+This repo is built upon [LLaMA-Adapter](https://github.com/OpenGVLab/LLaMA-Adapter).
+
+## Citations
+
+```
+@inproceedings{ko2023large,
+  title={Large Language Models are Temporal and Causal Reasoners for Video Question Answering},
+  author={Ko, Dohwan and Lee, Ji Soo and Kang, Wooyoung and Roh, Byungseok and Kim, Hyunwoo J},
+  booktitle={EMNLP},
+  year={2023}
+}
+```
+

checkpoint/dramaqa.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2563e5daa3614f31d8e8806e7fdca9da96094811d2dcd6aa97ea6961d24df5fa
+size 54058095

checkpoint/nextqa.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6509d11e3651564bc1d30f51cabe5e50419b84610fed86524011825eca59fb4
+size 54058095

checkpoint/star.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de117ad5552a3f25257df644789e6432e6dae47c8f206230fc9b20a8a2203d4c
+size 54058095

checkpoint/tvqa.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:54bbe0547765d20262c579ba7e3f37f9b5e3a737e372d1dda9a9b82f7acbfede
+size 54055407

checkpoint/vlep.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0044346ea500cffeef940b38152b149f7fa9b1223279a5a235e49553ad1be92c
+size 54058095

data/dramaqa/AnotherMissOhQA_test_set.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d289df433b8890edd565912a0039be754c79d44280399dcc3eaf8714fb5645b
+size 2442248

data/dramaqa/AnotherMissOhQA_train_set.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:12a0d579824c5e5eaacba4aff4559547da6b375638d506220d32c4e83c49e5c3
+size 11454438

data/dramaqa/AnotherMissOhQA_val_set.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e0f84e48eb2f6141dbc28a73d8de6589ff7e101495aebd1222982e0fa8e4f7d
+size 2467499

data/dramaqa/clipvitl14.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c07a8eb4964a484184472efb0cdee53a873c7084cf09a00475d9571f8da33a21
+size 347845769

data/nextqa/clipvitl14.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d73ea6941a096d5497443b325e0c693ca5cb5e0e58db44db32aa614f37f80176
+size 366419999

data/nextqa/train.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51ae6e611fa58835f63a0beb325fe647bb197c7e320a31f0bb7a9cf969e082b2
+size 5578444

data/nextqa/val.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43198bdef8436b8d64a9b75d846b0987c10cbf94ebf4be325c4a4e54634d66b8
+size 814107

data/star/STAR_test.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:769f3033a71bfd55a5042a94b1463696d563ce37447addd42b65e208ee5678fc
+size 2591326

data/star/STAR_train.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9dc9b8fd6026154ea99cd76ae2975957dd787312d88bb35b9d42df674fc3e257
+size 411564765

data/star/STAR_val.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c88335ce19281d882cb9b47f8126296cddee864f4ad0cf99bebc9c0a8533097c
+size 60041967

data/star/clipvitl14.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd32069bd79014196aac82b9949ae226a9e72efb903bac3b8ee75083883453e0
+size 452328545

data/tvqa/clipvitl14.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26187056e6c93102b328cd1d30a7f327da3f4741d2a9a88fe27ce94e2e87b33d
+size 7700125181

data/tvqa/tvqa_subtitles.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b711a623a41cf4bc497b096d7cf836b3cadfb590318923dafd58f84440dac4c8
+size 23814412

data/tvqa/tvqa_test_public.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:abe4718477756e9d37832e03b30718ce9096e41c103fa8297072c503d0264287
+size 2724100

data/tvqa/tvqa_train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0eb4297aa82604d7844ef011b97a125be4805668283772c93729db96dbddc02
+size 45599736

data/tvqa/tvqa_val.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e85ca671592ad915f53dddfb643eec643efd63c291578c830e4e1e46ade0015c
+size 5694418

data/vlep/clipvitl14.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a29f10d540d29b8d4a38defffb21b00afe4820e71db9c24ff96eef48a66e6a58
+size 509358025

data/vlep/vlep_dev_release.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e6d12fcc25900220c95ca458a803ee03d87ebbfbb665c7817792da260fdd6c46
+size 1078762

data/vlep/vlep_subtitles.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a67f471a16ec582daa1b545bcd246fb269239741ebd28b716f20e206f2d6e18b
+size 11616660

data/vlep/vlep_test_release.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3b23f549fb3133c524798cacf77ed4601b226067a1437ece3ba37f40fccc539
+size 977615

data/vlep/vlep_train_release.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bdae52205775829db507693386b8d24807b3f708bae0f3eea14f52a96000e085
+size 4984854