thtrieu/darkflow

training my own data set with ']' output predict

Open

#300 建立於 2017年6月23日

在 GitHub 查看
 (7 留言) (0 反應) (0 負責人)Python (6,120 star) (2,102 fork)batch import
bughelp wanted

描述

hi I tried to train my own data set using cfg/tiny-yolo-voc.cfg tiny-yolo-voc.weight. I have changed the class to 2, lable.txt, and the last conv filter to 35(5*(2+5)) the training script is as follow: ./flow --model cfg/tiny-yolo-voc-1.cfg --load bin/tiny-yolo-voc.weights --train --annotation train/Annotations --dataset train/JPEGImages --gpu 1.0 --batch 30 --epoch 100 after trained the loss is under 2.0 the testing script is: /flow --imgdir test_dir/ --model cfg/tiny-yolo-voc.cfg --load -1 --json the result is ']' why is it ? the sample under test data is just a training data.

貢獻者指南

技術棧
pythontensorflow
領域
machine learning
議題類型
bug
難度面向新貢獻者的預計實作難度，1 表示很小改動，5 表示專家級工作。
3
預計時間有經驗貢獻者完成調查、實作、測試並準備 pull request 的粗略時間範圍。
half day
活動狀態議題目前的可參與程度：新鮮、活躍、陳舊、阻塞或等待維護者輸入。
stale
清晰度議題是否清楚說明預期改動、驗收標準和下一步。
unclear
前置要求
Basic knowledge of YOLOFamiliarity with darkflowPython and TensorFlow setup
新手友善度1-100 的估計分數，表示該議題對首次貢獻者的友善程度。
30
研究方向
Investigate why the testing output is ']'. This likely indicates a parsing error or empty predictions. Check the JSON output format in darkflow's predict function (e.g., in 'darkflow/net/help.py' and 'darkflow/cli.py'). Verify that the cfg file modifications (changing classes and filters) are correct and consistent between training and testing. Test with a simple image to confirm model loading and forward pass. Look at the training script to ensure the model is properly saved and loaded.