Statistical Analysis Stack Exchange
Statistical Analysis Stack Exchange

Questions tagged [caffe]

Caffe is a deep learning framework. One of its main usages is the classification of image data.

Caffe is a deep learning framework made with expression, speed, and modularity in mind. It is developed by Berkeley AI Research (BAIR) and by community contributors. It is open source, under a BSD license. It is written in C++, with a Python interface. Caffe is often used for the classification and cluster analysis of image data.

Check out the official homepage:

http://caffe.berkeleyvision.org/

and the github repository:

https://github.com/BVLC/caffe

16 questions
15
votes
1 answer

Epoch Vs Iteration in CNN training

There are a few discussions for Epoch Vs Iteration. Iteration is one time processing for forward and backward for a batch of images (say one batch is defined as 16, then 16 images are processed in one iteration). Epoch is once all images are…
batuman
  • 441
  • 1
  • 3
  • 10
3
votes
0 answers

Recognition the same object from different views

I have 33 classes (33 different objects). I need to recognize the object from any view of the object. Like a packet of potato chips, the packet has different appearance from different view (as shown in the attached images). I trained those…
batuman
  • 441
  • 1
  • 3
  • 10
2
votes
1 answer

What is the purpose or benefits of fully connected layer at the middle of Convolutional Network?

Is there any benefits to have FC layer at the middle of CNN network? For example, in this network, FC7 has kernel size is 1. What is the benefits of using kernel size 1 in this use? Those inception net, kernel size 1 is used for lower down the…
batuman
  • 441
  • 1
  • 3
  • 10
2
votes
1 answer

Reduction of Feature map size in Convolutional Neural Network

In CNN, the way we reduce the feature map size at layers is we use pooling. Pooling makes feature map size into half. For the following network, if I want a new layer with feature map size somewhere around the middle between 30 and 15, what could be…
batuman
  • 441
  • 1
  • 3
  • 10
2
votes
1 answer

Is there an exhaustive dataset of just grayscale images or a CNN model ( preferably a caffe or tensorflow model ) Pre-trained on grayscale Images?

I have a very limited dataset of around 12k grayscale images and wanted to know if there is a CNN model that I can use for fine tuning or an grayscale image dataset that can be used for pre-training. I dont want to use models trained on ImageNet as…
Ayush Pandey
  • 141
  • 1
  • 5
1
vote
1 answer

Why py-faster-rcnn has 0 detection in trained model with no pretrained weight?

I tested py-faster-rcnn and found that if I don't use pretrained weight then mAP is 0 in testing. I hope that I should see some value eventhough not very high. With pretrained model, mAP was above 99%. Why so difference and how can I solve the…
batuman
  • 441
  • 1
  • 3
  • 10
1
vote
1 answer

How to get validation accuracy in neural network training?

The following plot shows over fitting in training using training accuracy and validation accuracy during training. How to get that plot in training? My understanding is that we have training set and test set of data for training. Using test data…
batuman
  • 441
  • 1
  • 3
  • 10
1
vote
1 answer

GoogleNet Loss Skyrockets

I am training GoogleNet on the Stanford cars data set. It's 8000 training images of cars with labels (2004 Toyota Camry). I made minimal changes to the network. I just changed the loss outputs to 196 since I have 196 types of vehicles. I used the…
mj_
  • 233
  • 2
  • 5
1
vote
0 answers

Computation complexity and processing of one image for object detection in Convolutional Neural Network

How do I relate compute complexity in Convolutional Neural Network to processing time of one image in object detection for a given CPU/GPU's processing power? Say my CNN architecture needs 'parameters to tuned': 20984, 'compute_complexity': 0.039456…
batuman
  • 441
  • 1
  • 3
  • 10
1
vote
1 answer

Why training loss is not low even though mAP is quite high?

I have a small network as follow. I have 1400 trained images and 350 test images. I trained from scratch without any pretrained weight. Actually it is same as SSD and just that network architecture is changed. So loss function is also same as…
batuman
  • 441
  • 1
  • 3
  • 10
1
vote
1 answer

How many maximum classes can be recognized using VGGNet-16?

I have 200 classes to recognize. I prepare 600 images for each class and I have 120,000 number of images prepared to train. I am thinking to use VGGNet-16. Would it be possible to recognize such number of classes using VGGNet-16?
batuman
  • 441
  • 1
  • 3
  • 10
1
vote
1 answer

Regularization in VGGNet-16 Network

I am looking into VGGNet. The networks are structured using Conv, Relu and Pooling layers only. How regularization is done in the VGGNet?
batuman
  • 441
  • 1
  • 3
  • 10
1
vote
1 answer

Developing a new Convolution Neural Network from scratch

I like to develop a new convolution neural network from scratch and the network is shown at the bottom (only main trunk without data input and classifier). For that, I need a pretrained model and here are few options tutorial 1 and tutorial 2. I…
batuman
  • 441
  • 1
  • 3
  • 10
1
vote
1 answer

Does the image size for convolutional neural networks have to be small?

I'm new to Caffe. I'm trying to understand the rules of using convolutional neural networks. And my questions are: is it mandatory for the images to be small for training? is it mandatory for the images to be squared? I'm asking this because…
Amani
  • 58
  • 7
0
votes
0 answers

How much is expected loss in training using SSD?

Is there any expected loss in training using SSD? Say I train one class detection and use 1200 images in training. What would be my expected loss when training is optimum? We normally stop training when loss doesn't change much? How can I verify…
batuman
  • 441
  • 1
  • 3
  • 10
1
2