Fig. 1. Overall architecture of GI-GAN model

Fig. 2. Comparison of model training loss. (a) Generated loss; (b) discriminant loss; (c) position offset loss

Fig. 3. ADE results of models under different K values. (a) S-GAN predicting loss; (b) GI-GAN-NA predicting loss; (c) GI-GAN predicting loss

Fig. 4. Comparison of prediction trajectories of different models. (a)-(g) Correct graphs of GI-GAN model trajectory prediction; (h)(i) error graphs of GI-GAN model trajectory prediction

Fig. 5. Multiple reasonable trajectories of GI-GAN-NA model

Fig. 6. Multiple reasonable trajectories of GI-GAN model

Fig. 7. Multiple reasonably predicted trajectories of same scene. (a) Maintain the original speed and turn to the horizontal direction; (b) slow down and wait, then turn to horizontal direction; (c) direct turn to the left and accelerate