search

dulucas / Displacement_Field

Official implementation of paper "Predicting Sharp and Accurate Occlusion Boundaries in Monocular Depth Estimation Using Displacement Fields" (CVPR2020)
120 stars 8 forks source link

supervised or unsupervised? #7

Closed jilner closed 4 years ago

jilner commented 4 years ago

Is this method supervised？ Do we need real ground truth of depth as the surveillance information?

dulucas commented 4 years ago

This method is supervised. In this repo, we use real depth as training data and blurring the depth image as our input. But, we also tried with synthetic depth and find the model trained with synthetic depth works just fine on real images. More interestingly, the model can be trained on synthetic images that are completely different from the depth images while still can maintain a certain level of performance. Hope it helps.

jilner commented 4 years ago

thx

---Original--- From: "Yuming Du"<notifications@github.com> Date: Thu, Oct 1, 2020 14:18 PM To: "dulucas/Displacement_Field"<Displacement_Field@noreply.github.com>; Cc: "jilner"<458994845@qq.com>;"Author"<author@noreply.github.com>; Subject: Re: [dulucas/Displacement_Field] supervised or unsupervised? (#7)

This method is supervised. In this repo, we use real depth as training data and blurring the depth image as our input. But, we also tried with synthetic depth and find the model trained with synthetic depth works just fine on real images. More interestingly, the model can be trained on synthetic images that are completely different from the depth images while still can maintain a certain level of performance. Hope it helps.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.