Error while converting to keras model about keras-yolo4

[elesun2018]

same problem

[farhantandia]

same problem with custom dataset

[Melkeydev]

@elesun2018 @farhantandia any updates? I am also on this issue with the original yolov4.weights downloaded/

[farhantandia]

the model is successfully converted, I forgot to specify the classes.txt if you are using custom dataset.

[Melkeydev]

@farhantandia I am using the stock yolov4.weight files?
When I run the convert.py, I am getting the buffer issue size. How did you resolve?

[farhantandia]

It also works for standard yolov4 weights in my case.

[Melkeydev]

@farhantandia can you share your code?
the convert.py - specifically? Still not working for me haha

[xellDart]

same problem, any solution?

[Shridharpawar77]

This convert file is also not working please provide working codes

[Ivan-basis]

@vihari1729 @xellDart which version of keras (and TF) do you use? I faced the same error due to another version of keras that has slightly different indexing mechanism of new created layers, i.e. in my case the first Conv2D layer had name "conv2d" (in keras 2.4.0) instead of expected "conv2d_1" (in keras 2.2.4). So, some layers become missed and weights are assigned by wrong names with shift to another layers. This leads to this error when it faces with another shape of some layer.

Try to use these versions

[benjamin2044]

In line 71(convert.py) change "conv2d_" to conv2D
if layer_name.startswith('conv2D'):

There will be warnings but the tensorflow model "yolo4_weight.h5" will be generated.