thedeepreader/detr_tutorial

detr_tutorial's Issues

Training the network from scratch

@thedeepreader Hi, I am training my network from scratch using resnet-50 backbone with imagenet pretrained weight on my custom dataset. The network is not converging, can you please help me in this regard. Below are my training details and plots of logs.

Backbone : resnet-50
learning_rate = 1e-4
learning_rate_backbone=1e-5
eos_coeff = 0.1
num_queries = 20
lerarning_rate_drop = 100

Extracting mAP for each class when evaluating on custom dataset

@thedeepreader I am using a custom COCO dataset.

Is there some way to extract mAP metric for each individual class/label when evaluating?

For example, I have 2 classes of the same object category. How do I address this in code to allow individual class results during inference?

Any help or advice is appreciated, thanks!

How are we going to change this part of the code if we have more than 1 class?

"
categories = [
{
"supercategory": "none",
"name": "face",
"id": 0
}
]

"

AssertionError "..detr\util\box_ops.py", line 51, in generalized_box_iou assert (boxes1[:, 2:] >= boxes1[:, :2]).all()

Data augmentation for custom training.

Dear Sirs,

Thanks to your tutorial, checkpoint file for my custom dataset was obtained.
To get more accuracy, I would like to add data augmentation during training.
From experience, "RGB to GREYSCALE" and "ROTATE 90" might be tried.

Would you please advice how to add above 2 augmentations.

Thank in advance.

How is the result?

Hello, I have trained the model in my machine, however my scores in log.txt seem incorrect, can you show me the training scores were trained under the correct?
My results are as flows, I have only trained 29 epoches because of the limitation of my machine.

Thank you! : )

How do I get the accuracy of DETR model?

error at line 157 datasets/face.py

function
make_Face_transforms(image_set) ---->make_face_transforms(image_set)