Adviser advice 就是 <font color="#f00"> Unet Accelerator </font> PMP security問peter Work Progress Update UNet的training flow已經由Jay建立起來，輸入圖片是224*224 gray scale picture。 UNet模型訓練完成。 模型參數已經dump出來，剩下每層之間的input activation還沒dump，目前仍在研究中-- by Leo。 接下來可能要延用之前的FPGA專題實作流程，將參數以及activation利用GEMM攤法，將資料攤成矩陣形式，只是需要更加robust。

目前還沒完工，但快了，還有一些小 bug 要解，以及 M 要用 fixed-point 運算的問題 inference_Unet_v1_8bit(ver1) # -*- coding: utf-8 -*- """ Created on Mon Sep 18 04:02:01 2023 @author: Jay """

Code # -*- coding: utf-8 -*- """ Created on Thu Aug 31 01:32:21 2023 @author: Jay """ # -*- coding: utf-8 -*-

由於我們需要知道 python 以及 Unet 背後真正的運算方式是什麼，因此要對 Unet 中有用到一些比較特別的 API 去做驗證以及理解 torchvision.transforms.functional.center_crop 會用在下圖中<font color="#999"> 灰色箭頭(copy and crop) </font>的運算，此 function 即是從圖片(from Contracting path of Unet)正中心向外擷取所需的圖片範圍，因此我們必須算出該從圖片的哪個 pixel 開始擷取並擷取至哪裡 </font> verification # -*- coding: utf-8 -*-

Congratulations! :tada: You have created a team workspace, where all team members collaborate on their shared notes. Team roles and note permission All members in the workspace have read access to every note in it. Only writers and admins have write access, and only admins can delete a team note. Team members receive notifications when someone else edit or mention them in a team note. Team members are all considered Owners to notes in the workspace when configuring note permissions.