Dear team,
> Only SODA10M dataset is allowed to be used for this Competition with one exception: the standard 1000-category ImageNet training set with its classification labels is allowed..
Q1. Does it mean that we are only allowed to use the ImageNet images but not the pre-trained weights ?
Q2. Could we use (any) open-source code without using pre-trained weights ?
Q3. Is it fine to use some rule-based approaches ?
Many thanks.
T
Q1. Does it mean that we are only allowed to use the ImageNet images but not the pre-trained weights ?
R1: Pre-trained weight (only with ImageNet dataset) is allowed to use.
Q2. Could we use (any) open-source code without using pre-trained weights ?
R2: You can use any open-source code, and pre-trained weights only with ImageNet dataset.
Q3. Is it fine to use some rule-based approaches ?
R3: Yes.
Best
Posted by: sslad2021 @ July 28, 2021, 3:40 a.m.Please clarify if COCO-pretrained object-detection models are allowed. Or only imagenet-pretrained encoders are allowed?
Posted by: dmitrykonovalov @ Aug. 8, 2021, 1:42 a.m.Q4: Please clarify if COCO-pretrained object-detection models are allowed. Or only imagenet-pretrained encoders are allowed?
R4: COCO-pretrained object-detection models is NOT ALLOWED, you can only use pre-trained weights with ImageNet dataset. Besides, you can only use ImageNet and SODA10M for the challenge.