TextBoxes-TensorFlow

TextBoxes re-implementation using tensorflow. Much more info can be found in Textbox and SSD This project is greatly inspired by slim project
And many functions are modified based on SSD-tensorflow project
Now the pipeline is much clear and can be resued in any tf projects.

Author: Daitao Xing : [email protected] Jin Huang : [email protected]

Result

For now, the model can detect most of boxes. But still has poor performance on small objects.

Train

Data preparation

1. download the sythtext data from link
2. Unzip the datasets and put all folders and files under /data/sythtext/
3. cd datasets and run script python data2record.py
4. The ICDAR2013 datasets follow the same pipeline.

Train from scratch

You can train this model from scratch by using following command.

DATASET_DIR=./data/sythtext/
TRAIN_DIR=./logs/train
CUDA_VISIBLE_DEVICES=0,1,2,3 setsid python Textbox_train.py \
	--train_dir=${TRAIN_DIR} \
	--dataset_dir=${DATASET_DIR} \
	--save_summaries_secs=60 \
	--save_interval_secs=1800 \
	--weight_decay=0.0005 \
	--optimizer=momentum \
	--learning_rate=0.001 \
	--batch_size=8 \
	--num_samples=800000 \
	--gpu_memory_fraction=0.95 \
	--max_number_of_steps=500000 \
    --use_batch=False \
	--num_clones=4 \

Train form checkpoint file

You can download vgg checkpoint from web and put it under checkpoints folder. To train the model from checkpoint(suggested), please run the following lines

# or change into your checkpoint file name
CHECKPOINT_PATH=./checkpoints/vgg_16.ckpt 
DATASET_DIR=./data/sythtext/
TRAIN_DIR=./logs/train
CUDA_VISIBLE_DEVICES=0,1,2,3 setsid python Textbox_train.py \
	--train_dir=${TRAIN_DIR} \
	--dataset_dir=${DATASET_DIR} \
	--save_summaries_secs=60 \
	--save_interval_secs=1800 \
	--weight_decay=0.0005 \
	--learning_rate=0.001 \
	--batch_size=8 \
	--num_samples=800000 \
	--gpu_memory_fraction=0.42 \
	--max_number_of_steps=500000 \
    --use_batch=False \
	--num_clones=4 \
	--checkpoint_path=${CHECKPOINT_PATH} \
    --checkpoint_model_scope=vgg_16 \
    --ignore_missing_vars=True \

Evalation on checkpoint

CHECKPOINT_PATH=./logs/train/logs614
EVAL_DIR=./logs/eval/logs614
DATASET_DIR=./data/ICDAR2013/test
CUDA_VISIBLE_DEVICES=4 setsid python eval.py \
    --eval_dir=${EVAL_DIR} \
    --dataset_dir=${DATASET_DIR} \
    --checkpoint_path=${CHECKPOINT_PATH} \
    --wait_for_checkpoints=True \
    --batch_size=1 \
    --gpu_memory_fraction=0.02 \
    --use_batch=False \

Further instruction

For further instruction on how to use slim. please check on slim project.

Problems to be solved：

1. Anchor boxes size is very important. Many small objects can't be matched with any anchor boxes. This is a drawback of SSD.
2. BatchNorm help a lot. Batch norm will imporove accurary by 5-10% percentage.
3. Tensorflow is ineffecient compared with Caffe. For now this pipiline is under optimization. The speed is about 3 sec/step on sigle gpu(K80) where batch size is 32.
4. Training as long as you can. I can't reproduce the result from scratch with only training 50K times. The result above is the result training after 150K