wyz

训练跑起来我也觉得很慢啊，好像测试时也很慢的

@corganhejijun Could you provided the trained model?Thanks!!!

@sunshinezhihuo,I run the ROLO successfully on the pretrained model,but I can't reproduce the result of the autor.The result run on the pretrained model is worse..

I modifiy the LSTM_Single function to with tf.device('/gpu:0'): # X, input shape: (batch_size, time_step_size, input_vec_size) # XT shape: (time_step_size, batch_size, input_vec_size) _X = tf.transpose(_X, [1, 0, 2]) # permute time_step_size...

你把这个路径改成你本地模型的路径就可以啦

The TensorFlow version of this project maybe 0.x,if you use newer version upper to 1.x, the api of the lstm have been changed.You should use the follow code for reference...

@hnyz979，你好！您提到的那个工程，看了下好像目前还没加上跟踪吧？我也是最近才接触到目标跟踪这个领域，希望能得到你的回复！

@DoubleWeiWei 我觉得，作者这个工程的ROLO是针对单目标跟踪的，MOLO才是针对多目标跟踪。然后单目标ROLO，其实他这个是用yolo作为特征提取网络，提取proposal，然后将yolo输出连续三帧的4096+6维特征输进lstm里面去做预测和跟踪，所以这里面不涉及到检测框的问题？？？不知道我这样理解的对不对？

@hnyz979 Hello!Can you send the "model_MOT.ckpt" file to my email [email protected]?? Thank you very much!!

I run the project on TensorFlow 1.5 by modified the LSTM_Single function to with tf.device('/gpu:0'): # X, input shape: (batch_size, time_step_size, input_vec_size) # XT shape: (time_step_size, batch_size, input_vec_size) _X =...