confuse about some code.

[tpys]

There are some code that I can't figure out by myself, please help!

1. in generator.py file, should the true_box_index be set to zero for every train_instance ?
   

   keras-yolo3/generator.py

   Line 68 in 25af1ba

   true_box_index = 0

2. in yolo.py file, pred_box_conf - 0 and true_box_wh + tf.zeros_like(true_box_wh) * (1-object_mask) make no difference, I really confused about warmup training, what it is trying to do ?
   please give more explanations.
   

   keras-yolo3/yolo.py

   Line 74 in 25af1ba

   conf_delta = pred_box_conf - 0

keras-yolo3/yolo.py

Line 153 in 25af1ba

true_box_wh + tf.zeros_like(true_box_wh) * (1-object_mask),

3. in YoloLayer there is a parameter named max_grid, when create model you pass the max_input_size to it, which is 480. but inside the create_yolov3_model this value has been multiply by 2 or 4 in different yolo layer. is that right?
   

   keras-yolo3/yolo.py

   Line 324 in 25af1ba

   [2*num for num in max_grid],

   
   

   keras-yolo3/yolo.py

   Line 348 in 25af1ba

   [4*num for num in max_grid],

   
   Thanks a lot !!!

[experiencor]

1. set to 0 every batch, not instance
2. 👍, my bad
3. yup, the grid size is doubled in size in later scales

[tpys]

@experiencor Thank you for kindly reply.

[muye5]

should 'max_grid ' be 480 or 480 / downsample (e.g. 480 / 32) . @experiencor @tpys

[artest08]

1. set to 0 every batch, not instance
2. +1, my bad
3. yup, the grid size is doubled in size in later scales

For the second one, is there an update. In the code still the line
true_box_wh + tf.zeros_like(true_box_wh) * (1-object_mask).

What should be this line? Should we add the width and height of the anchor boxes there?

I think the change can be done in this way:
true_box_wh + tf.ones_like(true_box_wh) *
self.anchors / net_factor *
(1-object_mask),