TypeError: buffer is too small for requested array about keras-yolo4

[Halkyon64]

Same problem here. Tried all possible mirrors for yolov4 weights.
But I've made only one change to convert.py: replaced from keras import backend as K to import tensorflow.python.keras.backend as K, because original import caused another problem:
AttributeError: module 'keras.backend' has no attribute 'get_session'

[Runist]

I meet the problem too.I search in Google,somebody say that it maybe your keras model isn't match yolov4.weights.So I try to figure out.

[Runist]

Traceback (most recent call last):
File "convert.py", line 173, in
yolo4_model = Yolo4(score, iou, anchors_path, classes_path, model_path, weights_path)
File "convert.py", line 157, in init
self.load_yolo()
File "convert.py", line 130, in load_yolo
buffer=weights_file.read(weights_size * 4))

I solve the problem,but I use my code.Here is link.Besides, my keras model is mismatch the AlexeyAB-version.So I just load csp-darknet.

[ybloch]

Got the same issue when I tried to convert my tiny-yolov4, trained with AlexeyAB darknet version

[JoshieLim]

I've got this error as well. Did anyone manage to solve it?

[Melkeydev]

any updates?

[JoshieLim]

@Amokstakov I've retrained my data by setting image size to (608,608) in yolov4-custom config file and it works.

[Melkeydev]

@JoshieLim I am struggling converting the downloaded yolo_4.weights file to the yolo4.h5 file using the conver.py
Any insight into this? I am lookign and the convert.py never actually references the .cfg file for yolo4?

[JoshieLim]

@Amokstakov Hmm that's odd. I'm able to convert to .h5 on my end. Did you set the anchors path to a yolo4_anchors? It should look something like this for yolov4 anchors 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401.

[Melkeydev]

Still no luck here.
Would it be too much trouble to ask for a direct step-by-step of what you did @JoshieLim

like from GitClone... what are the next steps into running the detect then test.py code?

[xellDart]

Any update?

[JoshieLim]

@Amokstakov Can you try to convert this yolov4.weights from this link. See if it works.

#get yolov4 pre-trained weights
!wget https://github.com/AlexeyAB/darknet/releases/download/darknet_yolo_v3_optimal/yolov4.weights

[MuhammadAsadJaved]

@ybloch Hi, Have you managed to convert tiny_yolov4.weights to .h5? It's pre-trained tiny_yolov4, not custom tiny_yolov4 model.

[ybloch]

@MuhammadAsadJaved Hi,
I could not convert it to H5 so I converted it to a TensorFlow model and with it, I use for production.
Here's how I did it:
Using repo: [https://github.com/hunglc007/tensorflow-yolov4-tflite]

• git clone https://github.com/hunglc007/tensorflow-yolov4-tflite

• cd tensorflow-yolov4-tflite

• sudo python3 save_model.py --weights yolov4-tiny-custom_final.weights --output ./checkpoints/yolov4_v1_darknet-416 --input_size 416 --model yolov4 --framework tflite

• sudo python3 convert_tflite.py --weights ./checkpoints/yolov4_v1_darknet-416 --quantize_mode fp16 --output ./checkpoints/yolov4_v1_darknet-416_fp16.tflite

You will find in this project an example of using the converted model

[MuhammadAsadJaved]

@ybloch OK. Thanks for sharing dear. Actually I need keras model for some other project (tiny_yolov4+deep_sort). It's a project for detection and then applying tracker on detections.

[Shridharpawar77]

did anyone manage to convert the model? If yes, please help

[maazullah96]

Thank @Runist for you comment I have converted yolov4.weights file to .h5 file using Runist YOLOV4 Repository "

[Mahmood-Hussain]

YOLOV4 Repository

@maazullah96 this shows 404

[Saqueeb]

Having same issue.
Did anyone convert custom trained yolov4 to .h5 model? If yes, then can you please share the steps here?