PaddleOCR/latest/quick_start #14035
giscus[bot]
bot
Replies: 4 comments 14 replies
-
|
一个项目的首页为什么连个docker或者正常的安装命令都找不到了 |
Beta Was this translation helpful? Give feedback.
-
|
收到反馈，已经逐步增加了，感谢。 |
Beta Was this translation helpful? Give feedback.
-
|
为啥训练完自己数据，转换为方便调用的形式，性能降了很多，是枝剪了很多吗？ 比如最表面的，训练完后的ocr的rec，模型有180M+大小，转换后，只有十几M大小。（比如：python tools/export_model.py -c "./configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml" -o Global.pretrained_model="./output/v3_en_mobile/iter_epoch_225.pdparams" Global.save_inference_dir="./inference_model/rec/"） |
Beta Was this translation helpful? Give feedback.
-
|
模型转换的应该有问题了，转的哪个模型呀
…---- Replied Message ----
| From | ***@***.***> |
| Date | 11/03/2024 21:33 |
| To | PaddlePaddle/PaddleOCR ***@***.***> |
| Cc | SWHL ***@***.***>,
Comment ***@***.***> |
| Subject | Re: [PaddlePaddle/PaddleOCR] PaddleOCR/latest/quick_start (Discussion #14035) |
为啥训练完自己数据，转换为方便调用的形式，性能降了很多，是枝剪了很多吗？ 比如最表面的，训练完后的ocr的rec，模型有180M+大小，转换后，只有十几M大小。（比如：python tools/export_model.py -c "./configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml" -o Global.pretrained_model="./output/v3_en_mobile/iter_epoch_225.pdparams" Global.save_inference_dir="./inference_model/rec/"）
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
|
大佬，还有点最后的疑问，大致就是②和④测试同样的数据，效果不一样。也用了单独转换前的识别部分，和单独转换后的识别部分对比。 以及检测识别 转换前后对比。 前后的会有区别，性能总体来说转换后的比转换前的差。 转换后的目的主要是更便于部署，是不是为了部署，枝剪了一些，导致性能下降。 |
Beta Was this translation helpful? Give feedback.
-
|
转换前后，据我说知应该没有做剪枝的。可以检查一下前后处理是否有出入呀 |
Beta Was this translation helpful? Give feedback.
-
|
大佬 请问OCR在构建数据集的时候应该按照什么格式加入负样本呢 |
Beta Was this translation helpful? Give feedback.
-
|
转换推理模型后，与预训练模型差距很大对吗。你可以检索以下问题库，推理模型预测和预训练模型预测逻辑不一样，有一个图片限制960你注释掉再试下 |
Beta Was this translation helpful? Give feedback.
This comment was marked as off-topic.
This comment was marked as off-topic.
-
|
在Ubuntu上运行时，发现占用的内存一直没有释放，导致跑几张图片后内存就占满了是为什么呢 |
Beta Was this translation helpful? Give feedback.
-
|
这个问题猜测是paddle框架推理前几张的图像的内存没有释放导致的。这个问题由来已久，可以试试更新最新版的paddle试试 |
Beta Was this translation helpful? Give feedback.
-
|
我试了最新版的3.0b2内存占用增加的速度变慢，但还是持续增加，并且识别速度大幅下降 |
Beta Was this translation helpful? Give feedback.
-
|
方便的话，提供一个可复现的最小demo吧，便于复现和排查。 |
Beta Was this translation helpful? Give feedback.
-
PaddleOCR/latest/quick_start
Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded and IoT devices)
https://paddlepaddle.github.io/PaddleOCR/latest/quick_start.html
Beta Was this translation helpful? Give feedback.
All reactions