search

YunYang1994 / tensorflow-yolov3

🔥 TensorFlow Code for technical report: "YOLOv3: An Incremental Improvement"
https://yunyang1994.gitee.io/2018/12/28/YOLOv3-算法的一点理解/
MIT License
3.63k stars 1.36k forks source link

How to train own images? #181

Open OevreFlataeker opened 5 years ago

OevreFlataeker commented 5 years ago

Hi, I want to try your code to try yolo with tensorflow. The provided examples work (Part 2: Quickstart), but I am not sure what to actually do with part 3. Why are you training on yolo voy in part 3.1 and not on custom images? WIth custom images I mean: train/test sets with a new object not former known to yolo, i.e. transfer learning (with new class)

I already have annotated pictures (both labelimg XML files as well as finished tfrecord files). How can I feed them into the process to let tf learn from them based on yolo? Also is there a tool which transforms the labelimg xml files into the format of dataset.txt shown in "Part 3"

So far I've used modesl/research/object_detection/model_main.py with the pipeline.config for training. How can I adapt this approach to yolo?

wszwedaEP commented 5 years ago

All you have to do is to prepare .txt files as described in README. Filename contains also filepath, so algorithm knows where to find the photo. After you've finished, adjust the config file in repository.

sourangshupal commented 4 years ago

i have done the labelling of 100 images using LabelImg. now i have 100 corresponding .txt files.

How to create the dataset.txt and anchor.txt file for custom dataset?