Is it possible to get full imagenet pretrained weights of MoCo v2 ?

[CharlieCheckpt]

❓ Is it possible to get full imagenet pretrained weights of MoCo v2 ?

Hello,
Thank you for this nice library.

I am trying to get the entire weights of resnet50 pretrained with moco v2 on ImageNet dataset. That is, I need the two encoders, the head etc ... so that I can resume the training with moco.

Is is possible to do this using the weights from the model zoo or somewhere else ?

I downloaded the weights and it does not seem to contain the weights of the two encoders.

[QuentinDuval]

Hello @CharlieCheckpt,

Thanks a lot for your question :)

Following your question, I checked the content of our model zoo for MoCo V2 and downloaded the weights available here:
https://dl.fbaipublicfiles.com/vissl/model_zoo/moco_v2_1node_lr.03_step_b32_zero_init/model_final_checkpoint_phase199.torch

Those weights are indeed complete and contain the weights of the trunk, the head, the momentum encoder, the queue, basically everything needed to restart training from there.

cp = torch.load("/path/to/model_final_checkpoint_phase199.torch")

# head
cp["classy_state_dict"]["base_model"]["model"]["heads"].keys()

# trunk
cp["classy_state_dict"]["base_model"]["model"]["trunk"].keys()

# loss (containing the momentum encoder)
cp['loss'].keys()

Please tell me if this answers your question,
Thank you,
Quentin

[CharlieCheckpt]

Hi @QuentinDuval , it does, sorry for some reason I couldn't find it.

Thank you !