Daniel Kusuma

VG zurück nach Hamburgo :D Thanks for the heads up. Enabling `use_07_metric` allows to calculate the mAP based of the precision v. recall curve using the 11 points interpolation. Not...

eval.py in this repo tells the model to run the detection, saves the result, and calculates the APs and the mAP. To see the detection result, you can do it...

I suppose they were benchmarking or trying to reproduce the original implementation on caffe, which trained the model on trainval set and evaluated it on the test set. I did...

I think that makes sense, since the training would usually take more time with lower learning rate, hence slower convergence. One of these approaches should improve the performance: normalising the...

I think in this case you have to either adjust the training labels or generate new training labels with only the desired classes

did you train your model from scratch or upload the pre trained weights on your model? maybe that's one thing to consider?

It refers to the mean of the ImageNet dataset. Since the dataset includes broad range of classes, these means are often applied to other similar dataset such as Pascal VOC...

Hi @umakrishnaswamy, thanks for the input! If I understood it correctly, I would then need to initialize the API object in as many processes as the number of forward pass...

@umakrishnaswamy yes, that would be very helpful. Thanks!

The input size after the preprocessing that you feed into the network has to have a fix size, 512x512 or 300x300 in the case of SSD300