ssd.pytorch
ssd.pytorch copied to clipboard
amdegroot
A PyTorch Implementation of Single Shot MultiBox Detector
As the **Table of Contents** We train the network by _train.py_ `python train.py` We evaluate the trained network by _eval.py_ `python eval.py` What puzzles me is that _test.py_ is used...
Do you still keep your loss info? Currently my loss keep hovers at 20. Can I know what loss value is considered reasonable?
When I was using visdom to display loss, I was just starting to have a picture, but when I refresh my browser, what happens when the picture goes away?
How can we train on fewer classes with the VOC dataset? I have changed the VOC_CLASSES dictionary to two labels and updated cfg['num_classes'] to 3, yet I still see the...
There is a mistake in line 129: `g_cxcy = (matched[:, :2] + matched[:, 2:])/2 - priors[:, :2]` where `priors[:, :2]` is wrong. `priors[:, 2:]` is correct.
If I interrupt the process of training, I can not reload the weight `ssd300_voc_10000.pth` which is the weight trained and saved in `weights/`, the error information is as follows. ```...
I tried to run many times with different parameters. But I could not achieve the reported mAP (over 77%). Almost mAP which I get are 71.x%. Could you tell me...
Hi. I was wondering what is the reason that the coco config 'num_classes' is 201 (in config.py), while coco has only 80 classes (in COCO_CLASSES, in coco.py)? Thanks.
I want to train on custom dataset with VOC format
> Looking at the source code in prior_box.py, we have code like pbox + = [cx, cy, ~, ~] in the forward function. If you analyze the loop with this...