List your company

Video Question Answering on TVQA

82.2AccuracyLLaMA-VQA
27.641.77555.9570.125Dec 9, 2022Feb 19, 2023May 2, 2023Jul 14, 2023Sep 24, 2023Dec 5, 2023Feb 16, 2024

Evaluation results

MethodLinks
LLaMA-VQALanguage Model=LLaMA, Number of Trainable Parameters=4.5M2023.1082.2
FrozenBiLM#PT=400M2022.1282
FrozenBiLMNumber of Frames=10, Speech Input=true2023.0582
VINDLU#PT=5M2022.1279
VINDLU#PT=25M2022.1279
VINDLU#PT=17M2022.1278.8
MERLOT#PT=180M2022.1278.7
MERLOTLanguage Model=ROBERTa, Number of Trainable Parameters=223M2023.1078.7
SPCRLLanguage Model=BERT2023.1076.2
VidLLanguage Model=BERT, Number of Trainable Parameters=25M2023.1076
HERO#PT=7.5M2022.1274.2
HEROInput=dense/1fps, Speech Input=true2023.0573.6
ViLAFrames Numbers=42023.1263.4
SEVILANumber of Frames=32 -> 4, Keyframe Selection=refined LOCALIZER2023.0561.6
SeViLAFrames Numbers=42023.1261.6
ViLAFrames Numbers=22023.1260.6
BLIP-2 concat (ANSWERER)Number of Frames=4, Protocol=feature concatenation2023.0559.8
SEVILA†Number of Frames=32 -> 4, Keyframe Selection=zero-shot LOCALIZER2023.0559.7
FrozenBiLMNumber of Frames=10, speech input or dense frames=true, Zero-shot=true2023.0559.2
FrozenBiLMNumber of Frames=102023.0557.5
FrozenBiLMFrames Numbers=102023.1257.5
HiTeA2023.1057.2
InternVideoNumber of Frames=8, Tested_by_authors=true2023.0557.2
InternVideoFrames Numbers=82023.1257.2
BLIP-2 votingNumber of Frames=4, Protocol=majority voting2023.0554.5
BLIP-2Frames Numbers=42023.1254.5
HGALanguage Model=GRU2023.1050.1
Q-ViDevaluation_mode=Zero-shot2024.0241
VideoChat2evaluation_mode=Zero-shot2024.0240.6
HCRNLanguage Model=LSTM, Number of Trainable Parameters=44M2023.1038.2
SeViLA†Number of Frames=32 -> 4, Zero-shot=true2023.0538.2
SeViLaevaluation_mode=Zero-shot2024.0238.2
BLIP-2concat (Answerer)Number of Frames=4, Zero-shot=true2023.0536.6
BLIP-2concatevaluation_mode=Zero-shot2024.0236.6
InternVideo*Number of Frames=8, Zero-shot=true2023.0535.9
InternVideoevaluation_mode=Zero-shot2024.0235.9
BLIP-2votingNumber of Frames=4, Zero-shot=true2023.0535.7
BLIP-2votingevaluation_mode=Zero-shot2024.0235.7
FrozenBiLMNumber of Frames=10, Zero-shot=true2023.0529.7
FrozenBILMevaluation_mode=Zero-shot2024.0229.7