Hi, is it acceptable to use pre-trained models such as a pre-trained vision encoder if it remains frozen (no optimization) during the entirety of its use? Thus it is treated as a static visual processing function which happens to be implemented as a CNN. Or is it the intention that any neural network used must be learnt from random initialization?
To clarify: it would be pre-trained on an unrelated dataset, such as ImageNet, and will have never been optimized using images from the simulator.
from the rules:
Participants should give as little information as possible to the robot, rather the system should learn from scratch […] given the difficulty of the competition and the many challenges that it contains and to encourage a wide participation, in Round 1 it will be possible to violate in part the aspects of the spirit of the competition, except the Golden Rule above. For example, it will be possible to use hardwired or pre-trained models for recognising the identity of objects and their position in space.
So you can use them for Round 1 but the submission wouldn’t be valid to advance to Round 2.
I would like to add that the competition is proving very hard at the moment, so even having a good score on Round 1 using pre-trained nets could be a very worthy submission.
To advance to Round 2 the pre-trained net should be then substituted by something else though, e.g. some autoencoder trained during the intrinsic phase.