wushaogui

Results 5 issues of wushaogui

Tensorrt is used to speed up reasoning. Generally, frozen Pb files are transformed into UFF or onnx files, and then into tensorrt reasoning engine. Can tfsecured encrypt tensorrt engine?

https://github.com/yjh0410/PyTorch_YOLOv1/blob/2335aee70380ea12606abe393f6782d28e0f84f9/utils/augmentations.py#L108 原文代码没有调整框的大小,经过数据增强后的图片,按照SSDAugmentation内的顺序去作数据增强,得到的图片框无法被准确画到目标上

obsidian-advanced-slides cannot render embedded blocks 1. This is the document to be embedded that I plan to use. One uses block to write and the other uses title to write...

bug

https://github.com/lyuwenyu/RT-DETR/blob/2b88d5d53bcbfbb70329bc9c007fdf7e76cf90dc/rtdetr_pytorch/src/zoo/rtdetr/rtdetr_postprocessor.py#L48 改为 `scores = F.softmax(logits, dim=-1)` 不然最后一个类别无法预测,因为结果被截断了,调试结果如下: ![image](https://github.com/lyuwenyu/RT-DETR/assets/24794718/fd3dfd29-8341-4950-a31a-ec6d93a57a6f)

token章节提出以下观点: ![image](https://github.com/user-attachments/assets/28f7b453-bec2-44af-82d3-7a0d2e60a302) 限制token=输入+ 输出 token 数之和,但是后文max_token=1,限制模型输出1个token ![image](https://github.com/user-attachments/assets/42adc67f-f5a9-44f8-93a9-b9211d2d65f4) 我不是专业的NLP工程师,请问这里该如何理解?