So I've had some luck either just changing them from jpg to png for some reason, cropping them better, or removing the background. But this may help develop some 3d characters more quickly. So of course this won't replace someone with a lot of skill in modeling and sculpting. Just click download and then you will have the 3d object. The finished files will be there and you can hover over to the right and three dots will appear. Open each of these so you have results > pifuhd_final > recon. When the computation is done, a new section will pop up called "results". png file with some type of normal map that I have no idea how to use, an. This can take a few minutes to run, so let it do it's thing.ġ0. This will automatically run all the code in sequence after the last one you ran. After this you can manually click on each bracket all the way down, or you can go up to the menu bar and click Runtime > Run After. If you already clicked on that, you can just click it again.ĩ. You want to change "test.png" to whatever the name of the file you just uploaded is. After you upload your image, go to the part of the code that has the " # example image" comment in green. So masking the background can help most of the time.ħ. The software tries to pick the person out, but sometimes it has trouble. You can crop them out or you can isolate the person and just have them on a white or transparent background. It's best to have a straight on shot if you can with a relaxed pose and no other people in the image. Click this and you can upload your own photo. You can hover to the right of "sample_images" and three dots will appear. Click on the folder icon on the very far left.
So uploading your own photo manually works better)ĥ. (Skip the "If you want to upload your own picture." part. After clicking on the "Configure input data" bracket section. The brackets will be next to some code to run.Ĥ. You have to skip over some text and video links in each section.
Scroll down and click on each of the brackets next to the code. At the top right click the pull down "Connect" > "Connect to Hosted Runtime"ģ. Click on the "Copy to Drive" button near the top. Make sure you're using Google Chrome and signed in to a Google account. There's a video showing how to use it, but the browser has an Asian language displayed. Luckily it can also be run in Google Collaborate right in the web browser. It's on GitHub and there is some way to compile it for Windows and Linux, but I think an Nvidia card is needed. It can take a 2d image and convert it to a pretty decent (most of the time) 3d model using neural networks. I saw someone post about this on Twitter.
0 kommentar(er)
