Скачать или смотреть TensorFlow 2 Custom Object Detection EP3 訓練

TensorFlow 2 Custom Object Detection EP3 訓練
帶領大家以最便利簡潔的方式來訓練Tensorflow, 並在Google Colab上直接訓練Model
訓練步驟：
1. Capture Pictures:
images\test
images\train
Edit labelmap.pbtxt

2. Download LabelImg:
https://github.com/heartexlabs/labelI...

3. Label Image:
產生 Xml

4. Generate TFRecord:
轉換照片 Xml - CSV - TFRecord
下載 generate_tfrecord.py
https://github.com/nicknochnack/Gener...
改名 generate_tfrecordCSV.py

open Anaconda Prompt
conda activate tf2
cd C:\Python_YT\TF_EP3
python generate_tfrecordCSV.py --labels_path=labelmap.pbtxt --xml_dir=images/test --output_path=test.record --csv_path=images/test_labels.csv
python generate_tfrecordCSV.py --labels_path=labelmap.pbtxt --xml_dir=images/train --output_path=train.record --csv_path=images/train_labels.csv

5. Model pipeline.config:
複製 模型至\TF_EP3\
編輯 pipeline.config:
num_classes: 2
train_config:
batch_size: 12
fine_tune_checkpoint: "efficientdet_d0_coco17_tpu-32/checkpoint/ckpt-0"
fine_tune_checkpoint_type: "detection"
train_input_reader:
label_map_path: "labelmap.pbtxt"
input_path: "train.record"
eval_input_reader:
label_map_path: "labelmap.pbtxt"
input_path: "test.record"

6. Train Model:
open Anaconda Prompt
conda activate tf2
cd C:\Python_YT\TF_EP3
copy C:\models-master\research\object_detection\model_main_tf2.py
python model_main_tf2.py --pipeline_config_path=pipeline.config --model_dir=training --num_train_steps=2000 --sample_1_of_n_eval_examples=1

7. Tensorboard:
評估模型:
cd C:\Python_YT\TF_EP3
python model_main_tf2.py --pipeline_config_path=pipeline.config --model_dir=training --checkpoint_dir=training

查看訓練狀況
tensorboard --logdir=training
http://localhost:6006

8. Export Model:
cd C:\Python_YT\TF_EP3
copy C:\models-master\research\object_detection\exporter_main_v2.py
python exporter_main_v2.py --trained_checkpoint_dir=training --pipeline_config_path=pipeline.config --output_directory=inference_graph

md outputs
python detect_from_image.py -m inference_graph\saved_model -l labelmap.pbtxt -i images\test

9. Google Colab:
https://github.com/BenGreenfield825/T...

step 1 Download Models
step 2 Compile Protocol Buffers & Install TensorFlow
step 3 Test whether TensorFlow is installed successfully
step 4 Download Model Zoo
step 5 測試物件偵測
step 5a Train Model
step 5b Tensorboard Train
step 6a 評估模型
step 6b Tensorboard Eval
step 7 Export Model


#TensorFlow #Detection #入門 #教學


00:00 TensorFlow 2 EP3 Training
00:13 Capture Pictures
01:07 Download LabelImg
02:27 Label Image
04:44 Generate TFRecord
06:23 Model pipeline.config
08:18 Train Model
09:19 Tensorboard
10:05 Export Model
11:24 Google Colab