MobilenetV2 SSDLite train finetune

1 Dataset:

   1.1 coco   http://cocodataset.org/dataset.htm#detections-challenge2017

  • python $WORKDIR/dataset_tools/create_coco_tf_record.py \
          --logtostderr \
          --train_image_dir="$DATADIR/train2014" \
          --val_image_dir="$DATADIR/val2014" \
          --train_annotations_file="$DATADIR/annotations/instances_train2014.json" \
          --val_annotations_file="$DATADIR/annotations/instances_val2014.json" \
          --output_dir="$WORKDIR/yx_project/dataset/COCO"
  • create coco_train.record 13G  with 82783 images
  • create coco_val.record 6.3G  with 40504 images

2 pre trained model:

   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_1.4_224.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_1.3_224.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_1.0_224.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_1.0_192.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_1.0_160.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_1.0_128.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_1.0_96.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_0.75_224.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_0.75_192.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_0.75_160.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_0.75_128.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_0.75_96.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_0.5_224.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_0.5_192.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_0.5_160.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_0.5_128.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_0.5_96.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_0.35_224.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_0.35_192.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_0.35_160.tgz
   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_0.35_128.tgz

   wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_0.35_96.tgz

3 build model/research/objectdetect/proto 

# From tensorflow/models/research/
protoc object_detection/protos/*.proto --python_out=.

Bug: 

1 TypeError: `pred` must be a Tensor, or a Python bool, or 1 or 0. Found instead: None

    Fix: change from mobilenet_v2.training_scope(is_training=None, bn_decay=0.9997))  

                        to     mobilenet_v2.training_scope(is_training=True, bn_decay=0.9997))

   

评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值