steps ref: https://medium.com/@kstseng/matterport-mask-rcnn-%E9%85%8D%E7%BD%AE-e62ced12c977
github: https://github.com/matterport/Mask_RCNN
為避免電腦各安裝包版本混亂，利用虛擬環境做區隔
教學：https://gist.github.com/Geoyi/d9fab4f609e9f75941946be45000632b
Environment:
python 3.5
cuda 9.1 cudnn 7.0.5
tensorflow-gpu 1.4.0 keras 2.0.8
scikit-image==0.13.1
coco dataset (download from yolo website script)
Run with jupyter notebook
enter samples: run demo.ipyb
訓練coco dataset (train2014 and val2014):
(p.s.
mrcnn/model.py - tf.reduce_mean(layer.output, keep_dims=True)
coco/annotations - add instances_minival2014.json, instances_valminusminival2014.json https://drive.google.com/drive/folders/0B1_fAEgxdnvJSmF3YUlZcHFqWTQ
train2014, val2014 拉出images folder
python samples/coco/coco.py train --dataset=../../yolo-v3/darknet/data/coco --model=coco
上圖為訓練時畫面
演算法筆記：https://blog.csdn.net/u014380165/article/details/81878644
安裝小紀錄先到這裡
ByeByeBye
Comments