dataloader
原有是图片对,现在应该取4张图片,一张作init,2张作为时序,另外一张作为searchimg。
首先修改videoanalyst/data/sampler/sampler_impl/track_pair_sampler.py
再修改videoanalyst/data/transformer/transformer_impl/random_crop_transformer.py,同时要改的是videoanalyst/data/utils/crop_track_pair.py
再修改videoanalyst/data/target/target_impl/densebox_target.py
trainer
videoanalyst/engine/trainer/trainer_impl/regular_trainer.py
forward
videoanalyst/model/task_model/taskmodel_impl/siamese_track.py