Closed ambikaverma closed 6 years ago
Hey! We use 2007+2012 to train in order to compare with the paper's results. If I recall correctly we achieve around 75.6 mAP on the 2007 test set as opposed to the 77.2 the authors quote on the paper for the same test set.
Hi
I saw the following blogpost - http://luminoth.readthedocs.io/en/latest/usage/checkpoints.html
which mentions an SSD model pre-trained on pascal dataset to be downloaded as a checkpoint and used. Could you share which pascal dataset is it trained on? Is it VOC 2007 or something else? Where can I find this information. Thanks!