Hi, I have successfully converted a darknet weights to params for mxnet but I have a problem of a missing json file? Do you guys have any idea how I could generate that file? I have used this method for converting the weights to params: https://github.com/bowenc0221/mxnet-yolo/tree/master/darknet2mxnet. Thanks guys.
It appears what you are linking does not contain yolo symbols: https://github.com/bowenc0221/mxnet-yolo/tree/master/Symbols it’s empty.
However good news, if you want to use Darknet for image classification you can use gluon-cv, and if you want yolo with Darknet it’s also available in gluon-cv.
Have a look at the following tutorial:
Hi, thanks for the response. However do you have a link for a none pretrained model as I want to train a YOLO weights from scratch?. What I did with my previous post is that I already have the converted weights that became params to be use with mxnet. Yesterday I was looking for this python code:
from gluoncv import model_zoo import mxnet as mx import numpy as np # Download the model from the Gluon model zoo # You'll find it in ~/.mxnet/models net = model_zoo.get_model('my_yolov3', pretrained=True) # Convert the model to symbolic format net.hybridize() # Build a fake image to run a single prediction # This is required to initialize the model properly x = np.zeros([1,3,224,244]) x = mx.nd.array(x) # Predict the fake image net.forward(x) # Export the model net.export('my_yolov3')
That block of code downloads a pretrained model that stores in ~/.mxnet/models/. Tried it and all I can see there is a model extended as .params. Now I looked deeper into the model_zoo.py script and since I have the converted param weights, all I needed is to load that weights locally. I wanted to ask if that is possible without having to download a pretrained weights from the internet but instead locally? Thanks…