Skip to content

Latest commit

 

History

History
47 lines (37 loc) · 2.83 KB

README_DETECTOR.md

File metadata and controls

47 lines (37 loc) · 2.83 KB
Raw

Object Detection

Current repository includes training and evaluation tools for general object-detection task. You can use one of the prepared config files to train the model on:

  • Pascal VOC0712 dataset (configs/detection/pascal_rmnet_ssd.yml)
  • MS COCO dataset (configs/detection/coco_rmnet_ssd.yml)
  • Pedestrian DB dataset (configs/detection/pedestriandb_rmnet_ssd.yml)

Data Preparation

To prepare a dataset, follow the instructions

Model Training

To train an object-detection model from scratch, run the command:

python2 tools/models/train.py -c configs/detection/pedestriandb_rmnet_ssd.yml \ # path to config file
                              -t <PATH_TO_DATA_FILE> \                          # file with train data paths
                              -l <PATH_TO_LOG_DIR> \                            # directory for logging
                              -b 4 \                                            # batch size
                              -n 1 \                                            # number of target GPU devices

NOTE: If you want to initialize the model from the pretrained model weights,specify the -i key as a path to init weights and set the valid --src_scope key value:

  • To initialize the model after pretraining on ImageNet classification dataset, set --src_scope "ImageNetModel/rmnet"
  • To initialize the model after pretraining on Pascal VOC or COCO detection dataset, set --src_scope "SSD/rmnet"

The command to run the training procedure from the pretrained model:

python2 tools/models/train.py -c configs/detection/pedestriandb_rmnet_ssd.yml \ # path to config file
                              -t <PATH_TO_DATA_FILE> \                          # file with train data paths
                              -l <PATH_TO_LOG_DIR> \                            # directory for logging
                              -b 4 \                                            # batch size
                              -n 1 \                                            # number of target GPU devices
                              -i <PATH_TO_INIT_WEIGHTS> \                       # initialize model weights
                              --src_scope "ImageNetModel/rmnet"                 # name of scope to load weights from

Model Evaluation

To evaluate the quality of the trained Object-Detection model, prepare the test data according to the instruction.

python2 tools/models/eval.py -c configs/detection/pedestriandb_rmnet_ssd.yml \ # path to config file
                             -v <PATH_TO_DATA_FILE> \                          # file with test data paths
                             -b 4 \                                            # batch size
                             -s <PATH_TO_SNAPSHOT> \                           # snapshot model weights