search

ijkguo / mx-rcnn

Parallel Faster R-CNN implementation with MXNet.
Other
671 stars 290 forks source link

ResNet101 as base net cannot detect any object? #86

Closed 315386775 closed 6 years ago

315386775 commented 6 years ago

Training stage.

INFO:root:Epoch[5] Batch [23280] Speed: 4.98 samples/sec Train-RPNAcc=0.995354, RPNLogLoss=0.012738, RPNL1Loss=0.234201, RCNNAcc=0.976568, RCNNLogLoss=0.063541, RCNNL1Loss=0.538337,
INFO:root:Epoch[5] Train-RPNAcc=0.995354 INFO:root:Epoch[5] Train-RPNLogLoss=0.012738 INFO:root:Epoch[5] Train-RPNL1Loss=0.234201 INFO:root:Epoch[5] Train-RCNNAcc=0.976568 INFO:root:Epoch[5] Train-RCNNLogLoss=0.063541 INFO:root:Epoch[5] Train-RCNNL1Loss=0.538337

Test stage

The AP == 0.00

Demo stage

I just changed the demo.py as : (1)from rcnn.symbol import get_resnet_test (2)symbol = get_resnet_test(num_classes=config.NUM_CLASSES, num_anchors=config.NUM_ANCHORS) Then run: python demo.py --prefix ./model/e2e --epoch 10 All result as the follow: results saved to ./reanetResult/000090.bmp class ---- [[x1, x2, y1, y2, confidence]]

How to solve this problem?

ijkguo commented 6 years ago

Note that in rcnn.config PIXEL_MEANS should be set to 0 for ResNet.

ijkguo commented 6 years ago

Closed due to inactivity.