Qualitative comparison — video versions

Each row is one held-out request, shown for all four methods at the same camera as the corresponding still in the paper. Videos are the motions the figures show as strobes.

Fig. 4 (main), row 1
Novel action combinations
“put down camera on chair, sit down on chair, pick up camera”
GT
Ours
ReMoGen
LINGO
Fig. 4 (main), row 2
Unseen rooms
“walk forward”
GT
Ours
ReMoGen
LINGO
Fig. 4 (main), row 3
Goals on unseen object categories
“sit down on chair”
GT
Ours
ReMoGen
LINGO
Supp. figure, group 1 row 1
Novel action combinations
“put down plant on chair, sit down on chair, pick up plant”
GT
Ours
ReMoGen
LINGO
Supp. figure, group 1 row 2
Novel action combinations
“pick up kettle, put down kettle, pick up cup”
GT
Ours
ReMoGen
LINGO
Supp. figure, group 1 row 3
Novel action combinations
“kneel down, pick up cup, drink from cup”
GT
Ours
ReMoGen
LINGO
Supp. figure, group 2 row 1
Unseen rooms
“put down guitar in right hand on shelf”
GT
Ours
ReMoGen
LINGO
Supp. figure, group 2 row 2
Unseen rooms
“stand up from seat”
GT
Ours
ReMoGen
LINGO
Supp. figure, group 2 row 3
Unseen rooms
“lie down on bed and face to right”
GT
Ours
ReMoGen
LINGO
Supp. figure, group 3 row 1
Goals on unseen object categories
“pick up plant with right hand from chair”
GT
Ours
ReMoGen
LINGO
Supp. figure, group 3 row 2
Goals on unseen object categories
“play guitar with both hands”
GT
Ours
ReMoGen
LINGO
Supp. figure, group 3 row 3
Goals on unseen object categories
“wash hands at washbasin”
GT
Ours
ReMoGen
LINGO