Reference

https://github.com/amdegroot/ssd.pytorch

What we do

We adapted the code to pytorch version 0.4.0. And some bug was fixed.

Still have some problems

Loading base network...
Initializing weights...
Loading the dataset...
Training SSD on: VOC2007
Using the specified args:
Namespace(basenet='vgg16_reducedfc.pth', batch_size=4, dataset='VOC', dataset_root='C:\\Users\\jimch/Documents/Github/ssd/data/VOCdevkit/', gamma=0.1, lr=0.001, momentum=0.9, num_workers=4, resume=None, save_folder='weights/', start_iter=0, visdom=False, weight_decay=0.0005)
timer: 7.9134 sec.
iter 0 || Loss: 27.4027 || timer: 0.1308 sec.
iter 10 || Loss: 14.9920 || timer: 0.1609 sec.
iter 20 || Loss: 20.7590 || timer: 0.1654 sec.
iter 30 || Loss: 20.2252 || timer: 0.1409 sec.
iter 40 || Loss: 15.1791 || timer: 0.1519 sec.
iter 50 || Loss: 14.1190 || timer: 0.1479 sec.
iter 60 || Loss: 13.9990 || timer: 0.1509 sec.
iter 70 || Loss: 120128.8672 || Traceback (most recent call last):
  File "train.py", line 224, in <module>
    train()
  File "train.py", line 149, in train
    loss_l, loss_c = criterion(out, targets)
  File "C:\Users\jimch\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 491, in __call__
    result = self.forward(*input, **kwargs)
  File "C:\Users\jimch\Documents\GitHub\ssd\layers\modules\multibox_loss.py", line 99, in forward
    _, loss_idx = loss_c.sort(1, descending=True)
RuntimeError: merge_sort: failed to synchronize: an illegal memory access was encountered

corresponding issue: amdegroot/ssd.pytorch#120