Shulou(Shulou.com)06/02 Report--

This article mainly talks about "what are the differences between CNN and RNN". Interested friends may wish to have a look. The method introduced in this paper is simple, fast and practical. Let's let the editor take you to learn "what are the differences between CNN and RNN"?

1. Comparison between CNN and RNN

1. Visual diagram of CNN convolution neural network and RNN recurrent neural network.

two。 Similarities:

The extension of traditional neural network.

The forward calculation produces the result and the reverse calculation model is updated.

Each layer of neural network can have multiple neurons coexisting horizontally, and there can be multi-layer neural network connection longitudinally.

3. Differences

CNN space expansion, neuron and feature convolution; RNN time expansion, neuron and multiple time output computation

RNN can be used to describe the output of continuous state in time, has memory function, and CNN is used for static output.

CNN Advanced 100 + depth, RNN depth is limited

2. CNN+RNN combination mode

1. CNN feature extraction, which is used to generate picture annotations in RNN statements.

2. RNN feature extraction is used for CNN content classification video classification.

3. CNN feature extraction for dialogue question and answer picture question and answer.

III. Specific application

1. Picture annotation

Basic ideas:

The goal is to generate annotated statements, a statement-generated task, LSTM?

Description of a large number of image information, image information expression, CNN?

The full connection layer feature description picture in CNN network, and the feature is combined with LSTM input.

Specific steps:

(1) Model design-feature extraction

The full connection layer feature is used to describe the original picture.

LSTM input: word+ image features; output the next word.

(2) Model design-data preparation

Image CNN feature extraction

Picture annotations generate Word2Vect vectors

Generate training data: picture feature + nth word vector: n + 1 word vector.

(3) Model training:

Use transfer learning, CNN features and sentence features to apply existing models

The final output model is LSTM, and the parameters of the training process are set: gradient upper limit (gradient clipping), learning rate adjustment (adaptivelearning).

It takes a long time to train.

(4) the model runs:

CNN feature extraction

CNN feature + the beginning of the sentence, word by word prediction

two。 Video behavior recognition:

What is happening in the video?

Summary of common methods:

(1) RNN is used for CNN feature fusion:

CNN feature extraction

LSTM judgment

Analysis of multiple recognition results.

Different features, different outputs.

Or: all features as one output.

(2) RNN is used for CNN feature filtering + fusion:

Not all video images contain definite classification information.

RNN is used to determine which frame is useful

For useful image feature fusion.

(3) RNN is used for target detection:

CNN directly generates the target candidate region

Fusion of candidate regions generated by LSTM pairs (the positions of adjacent moments are similar)

Determine the final exact location.

(4) Multi-model synthesis: in application, in order to produce * * results, multi-model ensemble is often used.

At this point, I believe you have a deeper understanding of "what are the differences between CNN and RNN". You might as well do it in practice. Here is the website, more related content can enter the relevant channels to inquire, follow us, continue to learn!

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.