Questiones about `decoder`

ZQPei / DSSD

Pytorch implementation of DSSD (Deconvolutional Single Shot Detector)

MIT License

56 stars 20 forks source link

Questiones about `decoder` #3

Closed FantasyJXF closed 4 years ago

FantasyJXF commented 4 years ago

The forward pass in DSSD/dssd/modeling/decoder/decoder.py, the features[-2-level] updates in each circle, did it do better than use the ORIGINAL feature comes from ResNet.

Another question, is this repo modified from mask-rcnn? The code architecture looks so familiar.

I'm really appreciate your work. Thank you very much.

FantasyJXF commented 4 years ago

Is there any pretrained model to test?

How's your evaluate results?

ZQPei commented 4 years ago

Thanks for telling me. Indeed, it is modified from mask-rcnn. The pretrained model is not released because I think it is not too hard to train. The mAP result on VOC is around 0.8 .

ZQPei commented 4 years ago

Hold on a second, I don't think there's anything wrong with my code. The features[-2-level] updates as x_conv for each loop and feed as x_deconv in the next one. This is exactly that is done in DSSD.

FantasyJXF commented 4 years ago

I didn't read original DSSD code, the picture on the paper seems didn't update x_conv during deconvolution module. By the way, I think this won't make much difference.

I port your code to ssd.prtorch, wait to see the results.

ZQPei commented 4 years ago

It would save memory.