uber-research / LaneGCN

[ECCV2020 Oral] Learning Lane Graph Representations for Motion Forecasting
https://arxiv.org/abs/2007.13732
Other
496 stars 131 forks source link

"left" and "right" not in gragh #34

Open hello-big-world opened 1 year ago

hello-big-world commented 1 year ago
for k1 in ["left", "right"]:
    graph[k1] = dict()
    for k2 in ["u", "v"]:
        temp = [graphs[i][k1][k2] + counts[i] for i in range(batch_size)]
        temp = [
            x if x.dim() > 0 else graph["pre"][0]["u"].new().resize_(0)
            for x in temp
        ]
        graph[k1][k2] = torch.cat(temp)

KeyError: 'left'