Skip to content

Latest commit

 

History

History
128 lines (91 loc) · 3.78 KB

usage_guide_mmdetection_v2.md

File metadata and controls

128 lines (91 loc) · 3.78 KB

Usage guide mmdetection 2.x

We will guide you on using mmdetection 2.x to train, test, and evaluate a model on laboro tomato dataset.

Installation

Requirements

To evaluate pretrained models please prepare mmdetection 2.x environment by official installation guide.

Prepare dataset

It is recommended to symlink the dataset root to $MMDETECTION/data. If your folder structure is different, you may need to change the corresponding paths in config files.

mmdetection
├── mmdet
├── tools
├── configs
├── data
│   ├── laboro_tomato
│   │   ├── annotations
│   │   ├── train
│   │   ├── test

Add datasets to mmdetection

To load data we need to create a new config file mmdet/datasets/laboro_tomato.py with corresponding subsets:

from .coco import CocoDataset
from .builder import DATASETS


@DATASETS.register_module()
class LaboroTomato(CocoDataset):
    CLASSES = ('b_fully_ripened', 'b_half_ripened', 'b_green', 
               'l_fully_ripened', 'l_half_ripened', 'l_green')

And add dataset names to mmdet/datasets/__init__.py:

from .laboro_tomato import LaboroTomato

__all__ = [    
           ..., 'LaboroTomato'
          ]

Configuration files

Configuration files setup on Tomato Mixed dataset example:

  1. Create laboro_tomato_base.py in configs/_base_/datasets/ with content of coco_detection configuration file and change dataset type, root and path parameters:
dataset_type = 'LaboroTomato'
data_root = 'data/laboro_tomato/'
...
  1. Create laboro_tomato_instance.py in configs/_base_/datasets/ with content of coco_instance and replace it with your base detection configuration file:
_base_ = 'laboro_tomato_base.py'
...
  1. Replace class numbers at model configuration file configs/_base_/models/mask_rcnn_r50_fpn.py:
...
num_classes = 6
...
num_classes = 6
...
  1. Replace dataset configuration file name in configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.py with created at step 3:
_base_ = [
    ...
    '../_base_/datasets/laboro_tomato_instance.py',
    ...
]

Evaluation

You can use the following commands to test a dataset:

# single-gpu testing
python tools/test.py configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.py \
                     laboro_tomato_96ep.pth --show

# multi-gpu testing with 4 GPUs
./tools/dist_test.sh configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.py \
                     laboro_tomato_96ep.pth 4 --out results.pkl --eval bbox segm

Train a model

To train your model finish all steps from Test a model section and change learning rate and total epoch, steps at configs/_base_/schedules/schedule_1x.py. The default learning rate in config files is for 8 GPUs and 2 img/gpu (batch size = 8*2 = 16). According to the Linear Scaling Rule, you need to set the learning rate proportional to the batch size if you use different GPUs or images per GPU, e.g., lr=0.01 for 4 GPUs * 2 img/gpu and lr=0.08 for 16 GPUs * 4 img/gpu.

optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001)
...
    step=[64, 88])
total_epochs = 96

You can use the following commands to train a model:

# single-gpu train
python tools/train.py configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.py \
                      --work-dir ./laboro_tomato

# multi-gpu train with 4 GPUs
./tools/dist_test.sh configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.py 4 \
                      --work-dir ./laboro_tomato