Rahul Somani

@rwightman After the brief hiccup in #606, I launched a run with the same hparams mentioned in the docs. It's quite interesting to see how EMA accuracy starts beating the...

Got it. That's super interesting.

@rwightman converged to 75.92 top-1 at epoch 515 (EMA). I've uploaded it to Dropbox [here](https://www.dropbox.com/s/033fu7vvsy1gn2q/checkpoint-515.pth.tar)

@rwightman I've included a url so that model weights are accessible with `pretrained=True`. As a sanity check, I ran the same model through the `validate.py` script, getting `Acc@1 75.924 (24.076)...

@rwightman no worries > You left the stem stride=2 un-blurred... was that intentional? That was an oversight... I have another training run for a separate task going on right now,...

Noted @rwightman. I haven't been able to launch the next run yet as the system's been occupied with some client project tasks. Given the run takes 4 days, I hesitate...

@rwightman I'm unsure what's the right approach towards downsampling the stem. If I understand correctly, the principle when applying blur pooling to a network is to ensure that the shapes...

@rwightman got it, in that case perhaps it makes sense _not_ to add anti aliasing to the mobilenet-v3 stem? ```python timm.create_model('mobilenetv3_large_100_aa') MobileNetV3( (conv_stem): Conv2d(3, 16, kernel_size=(3, 3), stride=(2, 2), padding=(1,...

+1 for this. Converting from PyTorch is possible but inference results don't match up exactly with Mask-RCNN (even with a batch size of 1), and it's hard to find some...

@rwightman that freezing code is a lot cleaner, I've updated mine to mimic the vision lock. Also implemented the free function as I understood it. Let me know if this...