1- Download Bosch Dataset, and combine with the cat command
2- Convert to TFRecord using the train and test script located in the google drive
python bosch_train_conversion.py && python bosch_test_conversion.py
3- Download the faster-rcnn-inception checkpoint from the object detection model zoo
https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md
4- Download faster-rcnn-inception-bosch.config file from drive
5- Change parameters in faster-rcnn-inception-bosch.config to your setup
batch_size, line 88
fine_tune_checkpoint, line 113, this points to the faster-rcnn-inception checkpoint gotten from step 3
input_path, lines 128 and 142
label_map_path, lines 130 and 144
6- Install object detection api
https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/installation.md
7- Start training bosch detector using
https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/running_locally.md
8- Download simulator images and labels from drive
9- Download label_map_simulator.pbtxt, simulator_conversion.py, faster-rcnn-inception-simulator.config from drive
10- Run TFRecord file conversion for simulator data
python simulator_conversion.py
11- Change parameters in faster-rcnn-inception-simulator.config to your setup
batch_size, line 88
fine_tune_checkpoint, line 113, this points to the faster-rcnn-inception-bosch checkpoint gotten from step 7
input_path, lines 128 and 142
label_map_path, lines 130 and 144
The model that is on the drive is this saved model
12- Freeze the simulator trained model
https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/exporting_models.md
13- Download trainer.py from the drive
14- Train the svm classifier using simulator data
python trainer.py
15- change to detection branch in our github, change the paths to the frozen detection model from step 12 and the svm model from step 14
adattudos
commented
6 years ago