Detection

This a Faster R-CNN model based on paper << Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks.>> http://arxiv.org/pdf/1506.01497.pdf written for learning purposes The program structure is base on Tensorflow's 'nmt'https://github.com/tensorflow/nmt and endernewton's workship tf-faster-rcnn https://github.com/endernewton/tf-faster-rcnn PS:For a good and more up-to-date implementation for faster/mask RCNN with multi-gpu support, please see endernewton's new example https://github.com/tensorpack/tensorpack/tree/master/examples/FasterRCNN

DataSet

MCOCO2014

Pre-trained model

download location:https://github.com/tensorflow/models/tree/master/research/slim VGG_16 ResNetV1_50 Inception_V3 unable to work temporarily,cause Tensorflow's avg pooling method don't supoort dynamic ksize and stride temporarily, see Issue tensorflow/tensorflow#26961

Environment

old: tensorflow-gpu: 1.13.1 CUDA: 10.0.130 CUDNN: 7.3.1 GPU: Nvida RTX-2070 new: tensorflow-gpu: 1.9.0 CUDA: 9.0.176 CUDNN: 7.3.0 GPU: Nvida RTX-2070

This model is unable to work now,cause (fixed)

UnknownError (see above for traceback): Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above. [[node vgg_16/conv1/conv1_1/Conv2D (defined at F:\Pycharm\PyCharm Community Edition 2018.3.5\workspace\Detection\models\vgg_16.py:20) ]] It looks like the error was caused by unmatching between CUDA and CUDNN, but i am not 100% sure. Solution Change environment to the new one may fix this problem

This model is unable to work now, cause the calculation from model is wrong when i draw calculated bounding box on images, remain fixing and updating