Centernet: demo fails detections with hourglass

[elinor-lev]

I run the demo with the pretrained ctdet_coco_hg.pth and it gives the trim warnings and fails to correctly find objects in the image. with a different pre-trained model ctdet_coco_dla_2x.pth it actually finds them correctly.

Running on Ubuntu 18.04, pytorch 1.4.0, CUDA 10.1

Result looks like this:

python demo.py ctdet --demo ../images/17790319373_bd19b24cfc_k.jpg --load_model ../models/ctdet_coco_hg.pth --debug 2
Fix size testing.
training chunk_sizes: [1]
The output will be saved to  /home/elinorm/codes/python/ImageDetection/CenterNet/src/lib/../../exp/ctdet/default
heads {'hm': 80, 'wh': 2, 'reg': 2}
Creating model...
loaded ../models/ctdet_coco_hg.pth, epoch 50
Drop parameter pre.0.conv.weight.If you see this, your model does not fully load the pre-trained weight. Please make sure you have correctly specified --arch xxx or set the correct --num_classes for your own dataset.
Drop parameter pre.0.bn.weight.If you see this, your model does not fully load the pre-trained weight. Please make sure you have correctly specified --arch xxx or set the correct --num_classes for your own dataset.
Drop parameter pre.0.bn.bias.If you see this, your model does not fully load the pre-trained weight. Please make sure you have correctly specified --arch xxx or set the correct --num_classes for your own dataset.
Drop parameter pre.0.bn.running_mean.If you see this, your model does not fully load the pre-trained weight. Please make sure you have correctly specified --arch xxx or set the correct --num_classes for your own dataset.
Drop parameter pre.0.bn.running_var.If you see this, your model does not fully load the pre-trained weight. Please make sure you have correctly specified --arch xxx or set the correct --num_classes for your own dataset.
Drop parameter pre.0.bn.num_batches_tracked.If you see this, your model does not fully load the pre-trained weight. Please make sure you have correctly specified --arch xxx or set the correct --num_classes for your own dataset.
Drop parameter pre.1.conv1.weight.If you see this, your model does not fully load the pre-trained weight. Please make sure you have correctly specified --arch xxx or set the correct --num_classes for your own dataset.
...
No param reg.0.bias.If you see this, your model does not fully load the pre-trained weight. Please make sure you have correctly specified --arch xxx or set the correct --num_classes for your own dataset.
No param reg.2.weight.If you see this, your model does not fully load the pre-trained weight. Please make sure you have correctly specified --arch xxx or set the correct --num_classes for your own dataset.
No param reg.2.bias.If you see this, your model does not fully load the pre-trained weight. Please make sure you have correctly specified --arch xxx or set the correct --num_classes for your own dataset.
tot 0.266s |load 0.008s |pre 0.011s |net 0.182s |dec 0.002s |post 0.063s |merge 0.000s |

[xingyizhou]

You need to specify the network architecture by --arch hourglass. The default architecture is DLA.