cupidjay / scaled-down-self-supervised-learning Goto Github PK

Hi,

Thanks for this great repo.

I was wondering if you could share the commands to reproduce results for BYOL on cars 196. Both for a resnet18 and resnet50.

Thanks!

Do you freeze the weight of the pre-training model based on the object detection? I checked your code. There is no blocking in the code.

I want to run your pre-train code on a single Colab GPU for my dataset. But I get the error: NotImplementedError: Only DistributedDataParallel is supported. My command is:

python main_moco_pretraining.py \
  -a resnet50 \
  --lr 0.03 \
  --batch-size 128 --epochs 200 \
  --input-size 224 \
  --gpu 0 \
  [path to my dataset]

Can you tell me how I can solve it, please?

Hi,
thank you for the repository!

I had a question: did you try fine-tuning models using SSL from models initialized from Imagenet.

Resnet50 pretrained on Imagenet --> fine-tune with MoCo on CUB-200 --> fine-tune for classification on CUB-200

I hope this is clear.

Thank you for any help!
Elias