ROLO outputs are always wrong, on the upper left corner

[chenharel1]

Hi guys, I ran the model and the code on a few given sequences, and I see that always the ROLO outputs are wrong, they are very small bounding boxes located in the upper left corner of the frames. I started debugging it, I suspect that the problem is in the initialization. I print self.pred_location (by: self.pred_location = tf.Print(self.pred_location,[self.pred_location]) And get very small values from the beginning (For example: [-0.0150853144 -0.0391756557 0.0126847308 0.0464992523], which basically is a small bb, a few pixels wide and height, close to the upper left pixel) I think that the first assignment actually ruins the whole process. Anyone has similar thoughts?