Saif Khan

ResNet-50 has 4.12 GFLOPs (i.e., 4.12 billion floating-point operations). See: [https://mmclassification.readthedocs.io/en/latest/model_zoo.html](https://mmclassification.readthedocs.io/en/latest/model_zoo.html). I get this exact value when I take `resnet50` from `torchvision.models` and calculate its FLOPs using this library.

But wouldn't it be better if the results were reported in a similar way to TensorFlow website? It is my understanding that it is common practice for most human pose...

@matthijsvk You can save the graph as SVG (or any other [format supported](https://graphviz.org/docs/outputs/) by `graphviz`) like this: ```python graph = draw_graph(model, ...).visual_graph graph_svg = graph.pipe(format='svg').decode('utf-8') # convert to binary data...

The closest data splits I found are these: https://gitlab.com/hzxie/Pix2Vox/-/blob/master/datasets/ShapeNet.json This file is from the official source code of [Pix2Vox++: Multi-scale Context-aware 3D Object Reconstruction from Single and Multiple Images](https://arxiv.org/abs/2006.12250) (IJCV...

This happens because of the following line in the implementation of `MultiDatasetEvaluator` class: https://github.com/open-mmlab/mmpose/blob/71ec36ebd63c475ab589afc817868e749a61491f/mmpose/evaluation/evaluators/mutli_dataset_evaluator.py#L39 Each sub-dataset in the `CombinedDataset` inside your validation set is expected to have a unique name....

According to the [PyTorch3D documentation](https://pytorch3d.org/docs/datasets), ShapeNetCore.v1 were rendered in this paper. Also, considering that V2 was [released in Fall 2016](https://shapenet.org/resources/releases/shapenetcorev2/README.txt), and this paper was published in ECCV'16 whose deadline was...