Talk2Nav: Long-Range Vision-and-Language Navigation with Dual Attention and Spatial Memory

Table 4 Comparison of our method to other methods at varying difficulty levels: from short paths consisting of one landmark to long paths consisting of four landmarks

Methods / #(Landmarks)	1	2	3	4
Random walk	31.05	16.95	9.04	2.88
Student-forcing (Anderson et al. 2018)	55.63	26.76	18.31	8.77
RPA (Wang et al. 2018)	59.75	30.30	18.20	8.43
Speaker-follower (Fried et al. 2018)	71.14	35.79	21.47	9.02
Self-monitoring (Ma et al. 2019)	72.21	37.71	23.16	9.56
Ours	74.28	43.08	27.96	11.92

SPL\(\uparrow \) is used as the metric
Boldness for the numbers in the tables signify that the corresponding row/method in the table gives the best performance among all other methods