ProGamerGov

I sucessfully trained a model that is similar to NIN but with less layers and produced the following images after training it for 70,000 iterations: https://imgur.com/a/sYRhV I used the CIFAR10...

@htoyryla As far as I understand, fine-tuning an already trained model means that you can use a smaller data set. So I have this data set here with art images:...

So I tried to fine-tune the VGG16 SOD model on the CIFAR10 data set, and received the following error: `ubuntu@ip-Address:~/caffe$ ./build/tools/caffe train -solver models/vgg16_finetune/solver.prototxt -weights models/vgg16_finetune/VGG16_SOD_finetune.caffemodel -gpu 0` ``` I0726...

I took the Cubo-Futurism jpg files from the people art data set. I then tried and failed to successfully create the val and train lmdb files.

So I have my images at: ``` /home/ubuntu/caffe/data/People-Art/JPEGImages/Academicism /home/ubuntu/caffe/data/People-Art/JPEGImages/AnalyticalRealism /home/ubuntu/caffe/data/People-Art/JPEGImages/ArtDeco /home/ubuntu/caffe/data/People-Art/JPEGImages/ArtNouveau(Modern) /home/ubuntu/caffe/data/People-Art/JPEGImages/Biedermeier /home/ubuntu/caffe/data/People-Art/JPEGImages/cartoon /home/ubuntu/caffe/data/People-Art/JPEGImages/Classicism /home/ubuntu/caffe/data/People-Art/JPEGImages/Constructivism /home/ubuntu/caffe/data/People-Art/JPEGImages/Cubism /home/ubuntu/caffe/data/People-Art/JPEGImages/Cubo-Futurism /home/ubuntu/caffe/data/People-Art/JPEGImages/Divisionism /home/ubuntu/caffe/data/People-Art/JPEGImages/EnvironmentalArt /home/ubuntu/caffe/data/People-Art/JPEGImages/FantasticRealism /home/ubuntu/caffe/data/People-Art/JPEGImages/FeministArt /home/ubuntu/caffe/data/People-Art/JPEGImages/HighRenaissance /home/ubuntu/caffe/data/People-Art/JPEGImages/Impressionism /home/ubuntu/caffe/data/People-Art/JPEGImages/InternationalGothic /home/ubuntu/caffe/data/People-Art/JPEGImages/Japonism /home/ubuntu/caffe/data/People-Art/JPEGImages/LowbrowArt /home/ubuntu/caffe/data/People-Art/JPEGImages/MagicRealism /home/ubuntu/caffe/data/People-Art/JPEGImages/MechanisticCubism etc... ```...

> I usually write small python scripts to manipulate or create the txt files in the correct format. https://stackoverflow.com/questions/11003761/notepad-add-to-every-line I just used this trick to fix my train and val...

The changes to my create_imagenet_2.sh file, val.txt, train.txt: https://gist.github.com/ProGamerGov/8267d29262f1bd6570e5918719600695 Still result in the same error.

@htoyryla Thanks, I'll make the modifications to my train_val.prototxt.

For the labels, do I put it as a different number value for each category?

Ok, I think I got it now. Change the `fc8_168` to `fc8_43` because I have 43 categories. Then change it to `fcpa_43`. Even with scripts in Notepad, it will take...