Spaces:

Zitang
/

Self-attention-based-V1MT-motion-model

Running on Zero

sunana commited on Dec 3, 2023

Commit

4679ee3

•

1 Parent(s): 4a13402

Update FFV1MT_MS.py

Files changed (1) hide show

FFV1MT_MS.py CHANGED Viewed

@@ -253,7 +253,7 @@ class FFV1DNN(nn.Module):
             assert len(flows_bi) == len(flows_up)
             results_dict["flow_seq"] = flows_up
-        flows_up = flows_up[:-2]
         attn_viz = attn_viz
         print(len(flows_up), len(attn_viz))

             assert len(flows_bi) == len(flows_up)
             results_dict["flow_seq"] = flows_up
+        flows_up = flows_up[:-1]
         attn_viz = attn_viz
         print(len(flows_up), len(attn_viz))