update scene_text_recognition/README.en.md#908
update scene_text_recognition/README.en.md#908liujiarik wants to merge 2 commits intoPaddlePaddle:dev-staticfrom
Conversation
| @@ -0,0 +1,141 @@ | |||
| To run the codes in this directory, we need to use v0.10.0 version. If your PaddlePaddle version is lower than this version, please update the PaddlePaddle according to the instructions in [installation document][1] | |||
There was a problem hiding this comment.
Running sample code in this directory requires PaddelPaddle v0.10.0 and later. If the PaddlePaddle on your device is lower than this version, please follow the instructions in [installation document][1] and make an update.
|
|
||
| # Scene Text Recognition(STR) | ||
|
|
||
| ## Introduction to scene text recognition task |
|
|
||
| ## Introduction to scene text recognition task | ||
|
|
||
| Many scene image contains rich text information, and it is very useful to know the content and meaning of the images. Therefore, scene text recognition is significant to learn Images. For example，the character recognition technology has promoted the development of the applications, such as: [[1][2]] . Which use deep learning to automatically identify signs of words , and help street view application to obtain more accurate address information. |
There was a problem hiding this comment.
STR plays a vital role in information acquisition of a scene image because the rich text information in a scene image may help us to better learn the content and meaning of the image. In addition, the development of STR has promoted the emergence of many new applications. For example, with deep learning models that automatically identify the texts in street signs, street view applications can acquire address information more accurately.
|
|
||
| Many scene image contains rich text information, and it is very useful to know the content and meaning of the images. Therefore, scene text recognition is significant to learn Images. For example，the character recognition technology has promoted the development of the applications, such as: [[1][2]] . Which use deep learning to automatically identify signs of words , and help street view application to obtain more accurate address information. | ||
|
|
||
| This example demonstrates how to complete the \* \* Scene Text Recognition (STR) \* \* task by PaddlePaddle. Task prepare a scene image ,which is shown in the figure below, `STR` need to identify the corresponding word "keep”. |
There was a problem hiding this comment.
an STR task with PaddlePaddle. We are tasked to identify the word "keep" shown in Figure. 1.
| </p> | ||
|
|
||
|
|
||
| ## Train and forecast by PaddlePaddle |
There was a problem hiding this comment.
Train and Forecast with PaddlePaddle
标题首字母大写。
| mv Challenge2\_Test\_Task3\_GT.txt data/test\_data | ||
| \`\`\` | ||
|
|
||
| 2.Get the path of the `gt.txt` in the training data folder (data/train\_data) and the path of the `Challenge2_Test_Task3_GT.txt` in the test data folder (data/test\_data). |
There was a problem hiding this comment.
Get the path of 'gt.txt'... and the path of Challenge2_Test_Task3_GT.txt
| --test\_file\_list\_path 'data/test\_data/Challenge2\_Test\_Task3\_GT.txt' \\ | ||
| --label\_dict\_path 'label\_dict.txt' | ||
| \`\`\` | ||
| 4.During training, the model parameters are automatically backed up to the specified directory, which is stored in the `./models` directory by default. |
There was a problem hiding this comment.
backed up to a speicified directory and stored under directory ./models by default.
|
|
||
| ### Matters needing attention | ||
|
|
||
| - Because the `warp CTC` of the model relies on the implementation of CUDA, this model only supports GPU operation. |
There was a problem hiding this comment.
Our model has to be runned by GPU since it relies on warp CTC, which can only be implemented by CUDA.
| ### Matters needing attention | ||
|
|
||
| - Because the `warp CTC` of the model relies on the implementation of CUDA, this model only supports GPU operation. | ||
| - The parameters of the model. The occupied memory is relatively large, the actual implementation can be adjusted by `batch_size` to control the memory usage. |
There was a problem hiding this comment.
With a lot of parameters, the model takes up a large amount of memory. When it is executed, its occupancy of the memory can be controlled by adjusting 'batch_size'.
|
|
||
| - Because the `warp CTC` of the model relies on the implementation of CUDA, this model only supports GPU operation. | ||
| - The parameters of the model. The occupied memory is relatively large, the actual implementation can be adjusted by `batch_size` to control the memory usage. | ||
| - The data set used in this example is small. If necessary, we can use another larger data set [[3][7]] to train the model. |
|
根据最新的格式要求，请将英文README文档的名称改为README.md。 |
|
liujia43 seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account. You have signed the CLA already but the status is still pending? Let us recheck it. |
4 participants
liujia43 pull request
fix #715