humphd / have-fun-with-machine-learning Goto Github PK

I've moved from Caffe/Digits to TensorFlow in my own work, and I want to update this to show how to do the same tasks with TensorFlow.

Given the hard work by @BirkhoffLee translating the original, I don't want to break what we already have. I'm debating whether to integrate it into the current document, or start a new one. I think it's nice to see how transferable the approach is across ML frameworks and technologies, so doing it in the same doc might make sense.

Thoughts?

root@8f88fbfbe029:/data/repo/test# python classify-samples.py -c /root/caffe/ -m epoch_30/ -d untrained-samples/
Traceback (most recent call last):
File "classify-samples.py", line 113, in
main()
File "classify-samples.py", line 95, in main
classifier = ImageClassifier(model_dir)
File "classify-samples.py", line 13, in init
self.net = caffe.Net(deploy_file, caffe.TEST, weights=weights_file)
Boost.Python.ArgumentError: Python argument types in
Net.init(Net, str, int)
did not match C++ signature:
init(boost::python::api::object, std::__cxx11::basic_string<char, std::char_traits, std::allocator >, std::__cxx11::basic_string<char, std::char_traits, std::allocator >, int)

Hello,

I like your language. it is written straightforward and easy to understand. actually that's an art if somebody has this ability. nice.

as you know GPUs are expensive, therefore do you know how we can train our model on a remote server with GPU cards (such as amazon) which we can rent hourly with low prices or similar?

Hello,

How did you noticed that which layers should be renamed in GoogLeNet prototxt?

I want to fine-tune the DetectNet but I don't know which layers should be renamed. The sample of object detection with Digits did not publish such part (they have not renamed anything).

Detectnet is not that much different with GoogLeNet so the fine tuning should not be that much different. I like your fine tuning because the classification model worked quite nice and accurately with even low number of images.

The job run for about 2 mins, but when on process#60 its crashing with the following error.see image attached.

Hello, thank you for your work!
Can I translate it into Korean?

Hi humphd,

Present i am trying to install Caffe and DIGITS using Dockers by following your steps.
I downloaded the Docker.

When i run this query on terminal
--> docker run --name digits -d -p 8080:5000 -v /path/to/this/repository:/data/repo kaixhin/digits

after all downloads i am getting error at last is -->
docker: Error response from daemon: Mounts denied:
The path path/to/this/repository is not shared from os x and is not known to docker.

Can you please tell me how to fix this.

Thank you.

原句
docker run --name digits -d -p 8080:5000 -v $(pwd)/have-fun-with-machine-learning:/data/repo /kaixhin/digits
是不是应该把kaixhin前的“/”去掉，改为
docker run --name digits -d -p 8080:5000 -v $(pwd)/have-fun-with-machine-learning:/data/repo kaixhin/digits

Hello，David Humphrey.
There are some question about the framework of image classification . You installed the OPNECL caffe. And you finished the image classification based on the Mac pro. You used the Inel graphics. But,the DIGITS cloud be installed if you would use a nvida GPU.
My laptop based on the AMD graphics , and i plan to run this work on VM(Ubuntu 16.04), It means that i can not install and use DIGITS ?
Thanks for reading.

Hi there,

I have cloned the repository to GitHub desktop and run the following command with the /path/to/this/repository replaced with my local directory of the repository.

docker run --name digits -d -p 8080:5000 -v /path/to/this/repository:/data/repo kaixhin/digits

My container exits immediately upon creation as seen when typing docker ps -a. Am I doing something wrong?

Thanks

Hi guys,
I hit an issue when I uploading the pertained models both Alex and google, like the below photos:

I have no idea about that, anybody can help me ? That would be a huge Appreciate.

Hello,

Consider we have some apples and we want to classify them. apples could differ in shape, size or sign of bruise on them.

Does deep learning classification could be able to distinguish these if we provide the related images for each class?, I mean something like this:

class-1: big apple
class-2: small apple
class-3: bad apple
class-4: good apple

Besides, consider we have an image which consists of three apples. Left, middle and right. how can I make the classifier to provide the classification results for these three apples as for example: left: good apple , middle: bad apple , right: big apple

Hello,

I followed everything and installed Caffe and Digits, but my Digits GUI is a bit different with yours. I mean there is no "Pretrained Models" and "processing" tab, and "segmentation" do not show up within drop-down list.

Why it should be like this?!

1.Digit Result

2.Python Result From This Tutorial

Here is my code

import numpy as np
import sys
import os

caffe_root = '/opt/caffe/'
sys.path.insert(0, os.path.join(caffe_root, 'python'))

import caffe
from caffe.proto import caffe_pb2

caffe.set_mode_gpu()
model_dir = 'model'
deploy_file = os.path.join(model_dir, 'deploy.prototxt')
weights_file = os.path.join(model_dir, 'snapshot_iter_64980.caffemodel')
net = caffe.Net(deploy_file, caffe.TEST, weights=weights_file)

transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})
transformer.set_transpose('data', (2, 0, 1))
transformer.set_raw_scale('data', 255)
transformer.set_channel_swap('data', (2, 1, 0))

mean_file = os.path.join(model_dir, 'mean.binaryproto')
with open(mean_file, 'rb') as infile:
blob = caffe_pb2.BlobProto()
blob.MergeFromString(infile.read())
if blob.HasField('shape'):
blob_dims = blob.shape
assert len(blob_dims) == 4, 'Shape should have 4 dimensions - shape is %s' % blob.shape
elif blob.HasField('num') and blob.HasField('channels') and
blob.HasField('height') and blob.HasField('width'):
blob_dims = (blob.num, blob.channels, blob.height, blob.width)
else:
raise ValueError('blob does not provide shape or 4d dimensions')
pixel = np.reshape(blob.data, blob_dims[1:]).mean(1).mean(1)
transformer.set_mean('data', pixel)
labels_file = os.path.join(model_dir, 'labels.txt')
labels = np.loadtxt(labels_file, str, delimiter='\n')

image = caffe.io.load_image('test_img.jpg')

net.blobs['data'].data[...] = transformer.preprocess('data', image)

out = net.forward()

softmax_layer = out['softmax']

LLPM_prob = softmax_layer.item(0)
OK_prob = softmax_layer.item(1)
YSBD_prob = softmax_layer.item(2)
YSCQ_prob = softmax_layer.item(3)

print(LLPM_prob)
print(OK_prob)
print(YSBD_prob)
print(YSCQ_prob)
`

My Python Result:

4.98672634421e-05
0.00573868537322
0.993777871132
0.000433590757893

Hello,

I decided to buy a GTX 1060 or GTX 1070 card to try with Deep Learning, but I am curious if the RAM size of The GPU or its bandwidth/speed will affect the accuracy of the final model or not, by comparing these two specific GPU cards.

in the other word, I want to know selecting the GTX 1060 will just cause longer training time over GTX 1070, or it will affect the accuracy of the model either.

Hey there. Thanks for the tutorial. Can I fork this repository and make a Chinese version of this? I really love this.

I have followed the exact steps from the tutorial. Tried with base learning rate 0.01 (default value) and 0.001 (based on the tutorial screenshot), but both didn't get better accuracy. Also it took almost 25 minutes to train.

Hi,
I'm using the docker option and it's really to setup. However, it's not using all the CPU the resources when training the model. It only use up to 100% cpu, which I believe is 1 core. But I gave the docker 12 cores.
Does any one know how to let it use more resource so it can run faster? Thanks

Hello,

First I should appreciate your tutorial.

My question is if we decided to detect objects in photos (draw rectangle on detected objects and get the coordinate of the center) what we should do?

and second, ho we can have video as input from the camera? (such as HDMi camera)