search

nwojke / deep_sort

Simple Online Realtime Tracking with a Deep Association Metric
GNU General Public License v3.0
5.33k stars 1.49k forks source link

Generate features for my own dataset #89

Open cppntn opened 6 years ago

cppntn commented 6 years ago

Hello,

How can i train a network to generate features for my own dataset (which is still made of person but not pedestrians)?

Moreover, is there a way to generate more than 128 features with the model you provided? Let's say i want to generate 512 features, how can i do that?

Thank you so much

aburney123 commented 6 years ago

To change the feature vector, you need to change the network architecture. Take a look at the architecture and modify/add layers to bring the output to 256. For example, one way of doing that is to change the output size of fully connected layer. The changes need to be made in the cosine-metric-learning/deep_sort/nets/network_definition.py

maylad31 commented 5 years ago

Hi,did you try increasing the size of the feature vector. Does this improve performance?