profile
viewpoint
Jeremie Papon jpapon Walt Disney Imagineering @disney Pasadena, CA http://www.jeremiepapon.com

jpapon/papon_thesis 2

My thesis -Done!

jpapon/cartographer 1

Cartographer is a system that provides real-time simultaneous localization and mapping (SLAM) in 2D and 3D across multiple platforms and sensor configurations.

jpapon/jpapon.github.io 1

Disputation Slide Show

jpapon/pcl 1

Point Cloud Library (PCL)

jpapon/apriltag2 0

A repository to start using the second version of the apriltag algorithm

jpapon/Azure_Kinect_ROS_Driver 0

A ROS sensor driver for the Azure Kinect Developer Kit.

jpapon/blensor 0

Blender Sensor Simulation

jpapon/caffe 0

Caffe: a fast framework for deep learning. For the most recent version checkout the dev branch. For the latest stable release checkout the master branch.

issue commentNVIDIA-AI-IOT/trt_pose

How can I train trt_pose on multi-GPU?

For reference, I just added: model = torch.nn.DataParallel(model) after the amp initialization in train.py and it uses multiple gpus. It looks like I'm not constrained by the data loader though, so I don't think that's being parallelized properly. @kinglintianxia did you do anything else for this?

kinglintianxia

comment created time in a month

issue openedNVIDIA-AI-IOT/trt_pose

Training non-square aspect ratios

Hello,

I've managed to get training up and running without issue as long as I keep image_shape square. If I try to change it something like [480,272] I get an error like the following: RuntimeError: The size of tensor a (128) must match the size of tensor b (64) at non-singleton dimension 2

Is it possible to train on non-square aspect ratios, and if so, what parameters do I need to adjust to do this? If not, what's the recommended way of using the network on input data that isn't square? Should I crop images to square, or does the network expect squished people because of training and I should just pass it a squished image?

Thanks!

created time in a month

issue commentNVIDIA-AI-IOT/trt_pose

Model description

+1 could you give more info on target_shape? In particular, if I train with a non-square aspect ratio, should I keep target_shape square, or adjust it so both dimensions remain 0.25*image_dim

Thanks!

michelazaccaria

comment created time in a month

PR opened NVlabs/imaginaire

Epoch save fix

Simple fix that corrects the save checkpoint behavior at the end of epochs - the current version checks iter.

+1 -1

0 comment

1 changed file

pr created time in 2 months

push eventjpapon/imaginaire

Jeremie Papon

commit sha baf4c03a2666838a454f42631e22cbe2d1ff9c6e

Fix for forgetting to check save start

view details

push time in 2 months

create barnchjpapon/imaginaire

branch : epoch_save_fix

created branch time in 2 months

fork jpapon/imaginaire

NVIDIA PyTorch GAN library with distributed and mixed precision support

http://imaginaire.cc/

fork in 2 months

issue commentNVlabs/imaginaire

Other Resolutions in SPADE

Excellent, thank you!

jpapon

comment created time in 2 months

issue commentNVlabs/imaginaire

Other Resolutions in SPADE

Training a new model with 512 or 1024 on the smallest side. The model seems to train fine, but the samples saved every image_save_iter come out blank.

Same data and configuration (other than the different smallest side and crop) trains and outputs sample images just fine at 256.

jpapon

comment created time in 2 months

issue openedNVlabs/imaginaire

Other Resolutions in SPADE

Hello,

Trying out the SPADE model in this library, and it works fine at 256x256. If I adjust the resize_smallest_side in base128_bs4.yaml to one of the other two values allowed in spade.py (512 or 1024), my generated images always turn out blank white.

Any idea why? Is there some other setting I need to adjust?

created time in 2 months

issue commentisl-org/MiDaS

3.0 models in ROS/LibTorch

Excellent @ranftlr thank you! I will test this on Friday.

jpapon

comment created time in 3 months

startedUKPLab/sentence-transformers

started time in 3 months

more