search

pjreddie / darknet

Convolutional Neural Networks
http://pjreddie.com/darknet/
Other
25.84k stars 21.33k forks source link

How can I use rnn to train model and classifiy image such as cifar10? #1446

Open blackwool opened 5 years ago

blackwool commented 5 years ago

Hi @pjreddie the rnn example in Darknet project is only to predict words . Can you upload examples training rnn network to classifiy image such as cifar10 or MNIST? or just paste rnn cfg file used to calssifiy MNIST ? Thank you very much .

AlexeyAB commented 5 years ago

@blackwool Hi,

Why do you want to use RNN to classify image cifar10/mnist instead of using convolution-networks (cifar.cfg, Tiny Darknet, AlexNet, ...) for this? The RNN(rnn, gru, lstm) usually can be applied to video (sequence of images) for tracking objects to slve the problem of objects occlusion.

blackwool commented 5 years ago

hi @pjreddie The darknet project is very small, smart and fast. I like it very much. I try to use RNN to classify video (sequence of images)action , but caffe and tensorflow are too big . The rnn_vid.c file gives me some hints that mybe darknet can do it.
But some files are no included in current darknet project such as extractor.cfg ,yolo_coco.con ,RNN model configure file, example video file and etc, and the program is not used for classify. So I come to you for help. May be you have think about it. Thank you very much.