From b7d77703f5caf76642ccd522ef83066181db3e13 Mon Sep 17 00:00:00 2001
From: wzyforgit <wangzhengyang@uniontech.com>
Date: Fri, 27 May 2022 16:22:58 +0800
Subject: [PATCH] =?UTF-8?q?fix:=20=E4=BF=AE=E5=A4=8D=E9=95=BF=E6=9D=A1?=
 =?UTF-8?q?=E5=9B=BE=E7=89=87=E6=97=A0=E6=B3=95=E6=A3=80=E6=B5=8B=E5=88=B0?=
 =?UTF-8?q?=E6=96=87=E6=9C=AC=20(#5)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

原因是当图片高度或宽度低于一定值后，检测算法难以执行检测
经过测试，图片的短边大于等于64的时候即可满足检测算法的要求
修改方法为在预先缩减图片尺寸后，进一步判断图片的短边是否符合大于等于64的要求，对不符合的执行第二次尺寸计算

Log: 修复长条图片无法检测到文本
---
 src/paddleocr-ncnn/details.cpp | 20 +++++++++++++++++---
 1 file changed, 17 insertions(+), 3 deletions(-)

diff --git a/src/paddleocr-ncnn/details.cpp b/src/paddleocr-ncnn/details.cpp
index 40c2a1c..75e1215 100644
--- a/src/paddleocr-ncnn/details.cpp
+++ b/src/paddleocr-ncnn/details.cpp
@@ -34,7 +34,7 @@ std::vector<std::vector<std::vector<int>>> Details::detectText(const cv::Mat &sr
     int w = src.cols;
     int h = src.rows;
 
-    //输入图片固定尺寸960*960
+    //1.缩减尺寸
     float ratio = 1.f;
     if (std::max(w, h) > 960) {
         if (h > w) {
@@ -47,9 +47,23 @@ std::vector<std::vector<std::vector<int>>> Details::detectText(const cv::Mat &sr
     int resizeH = int(h * ratio);
     int resizeW = int(w * ratio);
 
-    resizeH = std::max(int(round(float(resizeH) / 32) * 32), 32);
-    resizeW = std::max(int(round(float(resizeW) / 32) * 32), 32);
+    //2.扩大尺寸，确保图片的最短的边大于等于64,小于这个值可能会导致检测不到文字
+    float ratio2 = 1.f;
+    if (std::min(resizeW, resizeH) < 64) {
+        if (resizeH > resizeW) {
+            ratio2 = 64.0f / resizeW;
+        } else {
+            ratio2 = 64.0f / resizeH;
+        }
+    }
+    resizeH = int(resizeH * ratio2);
+    resizeW = int(resizeW * ratio2);
 
+    //3.标准化图片尺寸，否则会错位
+    resizeH = int(round(float(resizeH) / 32) * 32);
+    resizeW = int(round(float(resizeW) / 32) * 32);
+
+    //执行resize，记录变换比例
     cv::Mat resize_img;
     cv::resize(src, resize_img, cv::Size(resizeW, resizeH));
     float ratio_h = float(resizeH) / float(h);