Skip to content

fix(ocr): long text lines lost most characters (#3) #4

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
RGB-loop merged 1 commit into from
Oct 17, 2026
Merged

fix(ocr): long text lines lost most characters (#3) #4

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions src/core/ocr/index.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,9 +29,9 @@ let _modelsDir = ''
const remote = new RemoteReady()

/** 由 main 在启动时调用 */
/** v5 rec 模型对应的字典；换字典 / 模型版本时同时改 OCR_PIPELINE_VERSION，已有结果会重扫 */
/** v5 rec 模型对应的字典；换字典 / 模型版本 / 识别预处理时同时改 OCR_PIPELINE_VERSION，已有结果会重扫（v5-wide：长行不再被压到 320px） */
export const OCR_DICT_FILE = 'ppocrv5_dict.txt'
export const OCR_PIPELINE_VERSION = 'v5-dict'
export const OCR_PIPELINE_VERSION = 'v5-wide'

export function setOcrModelsDir(dir: string): void {
_modelsDir = dir
Expand Down
9 changes: 7 additions & 2 deletions src/core/ocr/recognition.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,10 @@ import type { RecognizedLine, TextBox } from './types'
// PaddleOCR v5 默认配置
const REC_IMG_H = 48 // 输入高度（v5 默认 48）
const REC_BATCH_SIZE = 6 // 每批一次性推理多少个文本框
const REC_MAX_WIDTH = 320 // 任一框最大输入宽度（防止极长行 OOM）
// 任一框最大输入宽度。PaddleOCR 按每行宽高比动态取宽、不设 320 的上限 —— 之前卡在 320（宽高比 < 7），
// 一行三四十个汉字被压成 320px、每字不到 10px，识别出来丢一大半（issue #3）。
// 3200 ≈ 宽高比 66，覆盖整行截图文字；更长的极端行才会被压缩（防 OOM：单框张量 3×48×3200 ≈ 1.8 MB）
const REC_MAX_WIDTH = 3200

// 归一化（PP-OCR v5 与 ImageNet 相同 mean/std；v3 是 [0.5, 0.5, 0.5]）
const MEAN = [0.5, 0.5, 0.5]
Expand Down Expand Up @@ -83,7 +86,9 @@ export async function recognizeBoxes(
}
if (crops.length === 0) return []

// 2) 分批：每批内 padding 到 max(w)，统一成 [B, 3, H, maxW]
// 2) 分批：按宽度排序（同 PaddleOCR 按宽高比排序）让每批宽度接近，padding 少；
// 每批内 padding 到 max(w)，统一成 [B, 3, H, maxW]。输出顺序无所谓，调用方按位置重排
crops.sort((a, b) => a.w - b.w)
const results: RecognizedLine[] = []
for (let i = 0; i < crops.length; i += REC_BATCH_SIZE) {
const batch = crops.slice(i, i + REC_BATCH_SIZE)
Expand Down
Loading