MMDetection starts fast and trains custom datasets

This article provides a quick guide to using MMDetection, documenting some of the issues to look out for in practice.

Environment to prepare

Based on the environment

Nvidia graphics card host
Ubuntu 18.04
- System installation, visible production USB boot disk, and system installation
Nvidia Driver
- Driver installation, visible Ubuntu initial configuration – Nvidia driver

The development environment

Download and install Anaconda, and then run the following command in Terminal:

Create a Python virtual environmentConda create-n open-mmlab python=3.7 -y conda activate open-mmlabInstall PyTorch with CUDAConda install Pytorch ==1.6.0 TorchVision ==0.7.0 CudatoolKit =10.2 -C Pytorch -y# installation MMCVPIP install MMCV - full - f https://download.openmmlab.com/mmcv/dist/cu102/torch1.6.0/index.html# installation MMDetection
git clone https://github.com/open-mmlab/mmdetection.git
cd mmdetection
pip install -r requirements/build.txt
pip install -v -e .
Copy the code

Pytorch ==1.7.0 multi-card training issues, refer to this Issue. Command Reference:

Conda install Pytorch ==1.7.0 TorchVision ==0.8.1 cudatoolKit =10.2 -C Pytorch -y PIP install mmcV-full -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.7.0/index.htmlCopy the code

More installation methods, visible official documentation:

MMDetection – Installation
MMCV – Installation

Existing models are extrapolated

Faster RCNN

Using r-50-FPN as an example, download its model file to MMDetection /checkpoints/. And then, by inference,

conda activate open-mmlab

cd mmdetection/

python demo/image_demo.py \
demo/demo.jpg \
configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py \
checkpoints/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth
Copy the code

Test existing models

Preparing the data set

Download the COCO dataset and place it in the mmDetection /data/ COCO/directory as follows.

├─ data │ ├─ coco │ ├─ annotations │ ├─ train2017 │ ├─ val2017 │ ├─ test2017Copy the code

Testing existing models

cd mmdetection/

# single-gpu testing
python tools/test.py \
configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py \
checkpoints/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth \
--out results.pkl \
--eval bbox \
--show

# multi-gpu testing
bash tools/dist_test.sh \
configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py \
checkpoints/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth \
2 \
--out results.pkl \
--eval bbox
Copy the code

The effect is as follows,

The results are as follows,

loading annotations into memory... Done (t = 0.33 s) creating the index... index created! [> > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > >] 5000/5000, 15.3 task/s, elapsed: 328 s, ETA: 0s writing results to results.pkl Evaluating bbox... Loading and preparing results... DONE (t = 0.89 s) creating the index... index created! Running per image evaluation... Evaluate annotationtype* Bbox * DONE (t=26.17s). Accumulating Evaluation Results with conditionals... DONE (t = 4.10 s). The Average Precision (AP) @ [IoU = 0.50:0.95 | area = all | maxDets = 100] = 0.374 Average Precision (AP) @ [ Ious = 0.50 | area = all | maxDets = 1000] = 0.581 Average Precision (AP) @ [IoU = 0.75 | area = all | maxDets = 1000] = 0.404 Average Precision (AP) @ [IoU = 0.50:0.95 | area = small | maxDets = 1000] = 0.212 Average Precision (AP) @ [IoU = 0.50:0.95 | Area = medium | maxDets = 1000] = 0.410 Average Precision (AP) @ [IoU = 0.50:0.95 | area = large | maxDets = 1000] = 0.481 Average Recall (AR) @ [IoU = 0.50:0.95 | area = all | maxDets = 100] = 0.517 Average Recall (AR) @ [IoU = 0.50:0.95 | area = All | maxDets = 300] = 0.517 Average Recall (AR) @ [IoU = 0.50:0.95 | area = all | maxDets = 1000] = 0.517 Average Recall (AR) @ [IoU = 0.50:0.95 | area = small | maxDets = 1000] = 0.326 Average Recall (AR) @ [IoU = 0.50:0.95 | area = medium | MaxDets = 1000] = 0.557 Average Recall (AR) @ [IoU = 0.50:0.95 | area = large | maxDets = 1000] = 0.648 OrderedDict ([('bbox_mAP', 0.374), ('bbox_mAP_50', 0.581), ('bbox_mAP_75', 0.404), ('bbox_mAP_s', 0.212), ('bbox_mAP_m', 0.41), ('bbox_mAP_l', 0.481), ('bbox_mAP_copypaste'.'0.374 0.581 0.404 0.212 0.410 0.481')])
Copy the code

Standard data set training model

Preparing the data set

Same COCO dataset as in the previous section.

Preparing configuration Files

The configuration file is configs/ faster_rCNN/faster_rCNn_R50_fpn_1X_coco-.py.

You need to modify LR learning rate parameters based on your GPU as follows:

Lr = 0.005 for 2 GPUs * 2 imgs/gpu
Lr = 0.01 for 4 GPUs * 2 imgs/gpu
Lr = 0.02 for 8 GPUs and 2 img/gpu (batch size = 8*2 = 16), DEFAULT
Lr = 0.08 for 16 GPUs * 4 imgs/gpu

_base_ = [
    '.. /_base_/models/faster_rcnn_r50_fpn.py'.'.. /_base_/datasets/coco_detection.py'.'.. /_base_/schedules/schedule_1x.py'.'.. /_base_/default_runtime.py'
]
# optimizer
optimizer = dict(type='SGD', lr=0.005, momentum=0.9, weight_decay=0.0001)
Copy the code

Training model

cd mmdetection/

# single-gpu training
python tools/train.py \
configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py \
--work-dir _train

# multi-gpu training
bash ./tools/dist_train.sh \
configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py \
2 \
--work-dir _train
Copy the code

Custom data set training model

Custom data sets

Here to demonstrate cat as a custom dataset from the Pascal VOC dataset,

conda activate open-mmlab

# Dataset Management Framework (Datumaro)
pip install 'git+https://github.com/openvinotoolkit/datumaro'
# pip install tensorflow

datum convert --input-format voc --input-path ~/datasets/VOC2012 \
--output-format coco --output-dir ~/datasets/coco_voc2012_cat \
--filter '/item[annotation/label="cat"]'
Copy the code

The data set needs to be in COCO format. Above, take CAT from VOC directly with datum and convert it to COCO format.

Preparing configuration Files

Add the configs/ VOC_cat/faster_rCNn_R50_fpn_1X_VOC_cat. py configuration file as follows:

# The new config inherits a base config to highlight the necessary modification
_base_ = [
    '.. /_base_/models/faster_rcnn_r50_fpn.py'.'.. /_base_/datasets/coco_detection.py'.'.. /_base_/schedules/schedule_1x.py'.'.. /_base_/default_runtime.py'
]

# We also need to change the num_classes in head to match the dataset's annotation
model = dict(
    roi_head=dict(
        bbox_head=dict(num_classes=1)))

# Modify dataset related settings
dataset_type = 'COCODataset'
classes = ('cat',)
data_root = '/home/john/datasets/'
data = dict(
    train=dict(
        img_prefix=data_root + 'VOC2012/JPEGImages/',
        classes=classes,
        ann_file=data_root + 'coco_voc2012_cat/annotations/instances_train.json'),
    val=dict(
        img_prefix=data_root + 'VOC2012/JPEGImages/',
        classes=classes,
        ann_file=data_root + 'coco_voc2012_cat/annotations/instances_val.json'),
    test=dict(
        img_prefix=data_root + 'VOC2012/JPEGImages/',
        classes=classes,
        ann_file=data_root + 'coco_voc2012_cat/annotations/instances_val.json'))
evaluation = dict(interval=100)

# Modify schedule related settings
optimizer = dict(type='SGD', lr=0.005, momentum=0.9, weight_decay=0.0001)
total_epochs = 10000

# Modify runtime related settings
checkpoint_config = dict(interval=10)

# We can use the pre-trained model to obtain higher performance
# load_from = 'checkpoints/*.pth'
Copy the code

modelconfigurationnum_classes=1Is the number of categories
datasetConfigure a custom data set for preparation
scheduleConfigured trainedlrAnd iteration roundstotal_epochs
runtimeconfigurablecheckpointHow many intervals to save one. The default is 1 epoch, and the space is insufficient 😶

The configuration can be overridden with __base__ content, as described in the official documentation.

Training model

# single-gpu training
python tools/train.py \
configs/voc_cat/faster_rcnn_r50_fpn_1x_voc_cat.py \
--work-dir _train_voc_cat

# multi-gpu training
bash ./tools/dist_train.sh \
configs/voc_cat/faster_rcnn_r50_fpn_1x_voc_cat.py \
2 \
--work-dir _train_voc_cat
Copy the code

When the breakpoint is restored,

bash ./tools/dist_train.sh \
configs/voc_cat/faster_rcnn_r50_fpn_1x_voc_cat.py \
2 \
--work-dir _train_voc_cat \
--resume-from _train_voc_cat/epoch_100.pth
Copy the code

ModuleNotFoundError: No module named ‘PycocoTools’

pip uninstall pycocotools mmpycocotools
pip install mmpycocotools
Copy the code

Viewing training Loss

pip install seaborn

python tools/analyze_logs.py plot_curve \
_train_voc_cat/*.log.json \
--keys loss_cls loss_bbox \
--legend loss_cls loss_bbox
Copy the code

See log.json for keys.

The test model

# single-gpu testingpython tools/test.py \ configs/voc_cat/faster_rcnn_r50_fpn_1x_voc_cat.py \ _train_voc_cat/latest.pth \ --out results.pkl \ -eval bbox \
--show

# multi-gpu testing
bash tools/dist_test.sh \
configs/voc_cat/faster_rcnn_r50_fpn_1x_voc_cat.py \
_train_voc_cat/latest.pth \
2 \
--out results.pkl \
--eval bbox
Copy the code

GoCoding personal practice experience sharing, please pay attention to the public account!

MMDetection starts fast and trains custom datasets

Environment to prepare

Based on the environment

The development environment

Existing models are extrapolated

Faster RCNN

Test existing models

Preparing the data set

Testing existing models

Standard data set training model

Preparing the data set

Preparing configuration Files

Training model

Custom data set training model

Custom data sets

Preparing configuration Files

Training model

Viewing training Loss

The test model

Related Posts

How can a bill be adopted more easily on Capitol Hill: AI tells you the trick

YOLOv5 target detection on Android

Machine Learning for Humans chapter 1: Why Machine Learning Matters