List your company

Video-to-command generation on standard object sets

0.618BLEU-1Human-to-Robot (Ours)
0.299760.382380.4650.54762Feb 22, 2026

Evaluation results

MethodLinks
Human-to-Robot (Ours)Backbone=ResNet-1012026.020.6180.5110.4030.351
Human-to-Robot (Ours)Backbone=ResNet-502026.020.6070.5060.3970.337
Human-to-Robot (Ours w/o key frame)Backbone=ResNet-1012026.020.5770.4650.3630.298
Human-to-Robot (Ours w/o key frame)Backbone=ResNet-502026.020.5620.4640.3610.297
Human-to-Robot (Ours w/o object sel.)Backbone=ResNet-1012026.020.5240.4110.3030.237
Human-to-Robot (Ours w/o object sel.)Backbone=ResNet-502026.020.5210.4070.30.237
Watch-and-ActBackbone=InceptionV32026.020.3990.2870.2630.199
Watch-and-ActBackbone=ResNet-502026.020.3940.2710.2480.187
V2CBackbone=ResNet-502026.020.3570.2310.2010.153
V2CBackbone=InceptionV32026.020.3550.2270.2010.164
Watch-and-ActBackbone=ResNet-1012026.020.3390.1710.1550.134
V2CBackbone=ResNet-1012026.020.3240.1630.1530.131
Video2CommandBackbone=InceptionV32026.020.3230.1960.1620.154
Video2CommandBackbone=ResNet-1012026.020.3190.1830.1750.148
Video2CommandBackbone=ResNet-502026.020.3120.1960.1730.15