Following up on #GeoTorchAI :

With a few minor changes, I managed to run the #bicycle flow prediction example of @geotorchai using #STResNet 🥳

The OD matrix size is rather modest (covering a grid with 21x12 cells), but hey, it's a start

https://github.com/anitagraser/GeoTorchAI/blob/f64cea879fdb846a2761b86ddefb5a946357b727/binders/prediction_spatiotemporal.ipynb

#geoai #resnet #torch #gischat #SpatialDataScience

Despite having worked with # PyTorch convolution operators, and their directly with their individual weights/parameters, for at least a year now, it's only now that I'm realizing that a PyTorch convolution operator can operate on channels of arbitrary size, so long as you feed it the expected number of input channels (and of course so long as the kernels are smaller than the images). Really clarifies some things about the #ResNet architectures I'm looking at.

A performance comparison of the new #PyTorch2 with the well established #PyTorch 1.

The benchmarks cover different areas of #deeplearning, such as image classification (#resnet) and language models (#BERT).

https://www.aime.info/blog/en/pytorch-2-gpu-performance/

A performance comparison of the new #PyTorch2 with the well established #PyTorch 1.

The benchmarks cover different areas of #deeplearning, such as image classification (#resnet) and language models (#BERT).

#мем #resnet #нейросеть #torch #cv #frustration

Este gajo inventa-me as ResNets, as Mask R-CNNs, as SPP-Nets, a PReLU, a inicialização He, co-autor da Focal Loss, das FPNs e muito mais...e tem mais de 340 mil citações.
Está claramente nas malhas do doping. Só não vê quem não quer.

#He #Kaiming #ResNet #Mask #R-CNN # PRelu #FocalLoss #FPN