view

[MYT999]

Why is my result like this? I followed the README step by step. Can you give me some advice? Thank you.

[MYT999]

Why is my result like this? I followed the README step by step. Can you give me some advice? Thank you.

(fast3r1) myt@MSI:~/桌面/fast3r1$ python fast3r/viz/demo.py
/home/myt/anaconda3/envs/fast3r1/lib/python3.11/site-packages/pl_bolts/init.py:11: FutureWarning: In the future np.object will be defined as the corresponding NumPy scalar.
if not hasattr(numpy, tp_name):
/home/myt/anaconda3/envs/fast3r1/lib/python3.11/site-packages/pl_bolts/models/self_supervised/amdim/amdim_module.py:34: UnderReviewWarning: The feature generate_power_seq is currently marked under review. The compatibility with other Lightning projects is not guaranteed and API may change at any time. The API and functionality may change without warning in future releases. More details: https://lightning-bolts.readthedocs.io/en/latest/stability.html
"lr_options": generate_power_seq(LEARNING_RATE_CIFAR, 11),
/home/myt/anaconda3/envs/fast3r1/lib/python3.11/site-packages/pl_bolts/models/self_supervised/amdim/amdim_module.py:92: UnderReviewWarning: The feature FeatureMapContrastiveTask is currently marked under review. The compatibility with other Lightning projects is not guaranteed and API may change at any time. The API and functionality may change without warning in future releases. More details: https://lightning-bolts.readthedocs.io/en/latest/stability.html
contrastive_task: Union[FeatureMapContrastiveTask] = FeatureMapContrastiveTask("01, 02, 11"),
/home/myt/anaconda3/envs/fast3r1/lib/python3.11/site-packages/pl_bolts/losses/self_supervised_learning.py:228: UnderReviewWarning: The feature AmdimNCELoss is currently marked under review. The compatibility with other Lightning projects is not guaranteed and API may change at any time. The API and functionality may change without warning in future releases. More details: https://lightning-bolts.readthedocs.io/en/latest/stability.html
self.nce_loss = AmdimNCELoss(tclip)
[13:17:25] ViserServerManager started demo.py:81
Loading weights from local directory

• Running on local URL: http://127.0.0.1:7860

Could not create share link. Please check your internet connection or our status page: https://status.gradio.app.
Extracted 9 frames from video using OpenCV

Loading a list of 9 images

• adding /home/myt/桌面/fast3r1/gradio_tmp_dir/8bd49782373be97a5ded1c5c5bbbca114bcd744f588d65e76b57a29297185c8f/frame_000000.jpg with resolution 1080x1920 --> 288x512
• adding /home/myt/桌面/fast3r1/gradio_tmp_dir/bb92478b1cfaf0453cb80139572db2fb7e712a51274ccf25377788f1093306b6/frame_000001.jpg with resolution 1080x1920 --> 288x512
• adding /home/myt/桌面/fast3r1/gradio_tmp_dir/49d28c275c89e3d61fd54bb973541d1afa6efadcad8646e97f830db3d6947dd9/frame_000002.jpg with resolution 1080x1920 --> 288x512
• adding /home/myt/桌面/fast3r1/gradio_tmp_dir/1506038ed69e9d1d153fc4976d326c012e7623ee87efc7225eee9362abe09626/frame_000003.jpg with resolution 1080x1920 --> 288x512
• adding /home/myt/桌面/fast3r1/gradio_tmp_dir/20c430bb70029e1a4e854c467067b548c93c82e19360b1c27f9a768fe73a6517/frame_000004.jpg with resolution 1080x1920 --> 288x512
• adding /home/myt/桌面/fast3r1/gradio_tmp_dir/43e83ec3c5196220cb0a4b6da600bc5b8575688f91efd1542f0fd1850957105c/frame_000005.jpg with resolution 1080x1920 --> 288x512
• adding /home/myt/桌面/fast3r1/gradio_tmp_dir/00d5a6ed55ca459406e80b523cb73035faa1a817ed530d440cc665d2c0861997/frame_000006.jpg with resolution 1080x1920 --> 288x512
• adding /home/myt/桌面/fast3r1/gradio_tmp_dir/4007ef0143d7e99057ffb85f6e299488f896e89851afd3aa4990ac854d76f57c/frame_000007.jpg with resolution 1080x1920 --> 288x512
• adding /home/myt/桌面/fast3r1/gradio_tmp_dir/6832e5c6ebd7333ad91b907d6d1de5f105552f4300210b7e63e0c57b628812ec/frame_000008.jpg with resolution 1080x1920 --> 288x512
  (Found 9 images)
  Image loading and cropping time: 0.28 seconds

Inference with model on 9 images
encode_images time: 0.3132355213165283
pos emb time: 0.0006678104400634766
decoder time: 0.19856023788452148
head prepare input time: 0.0008883476257324219
head forward time: 0.2701735496520996
total Fast3R forward time: 0.7836384773254395
[13:17:39] Launching viser server with id 1 for session demo.py:96
session_1741411059458 (message_id:
msg_1741411059458_37206_127101682218816)
╭──────────────── viser ────────────────╮
│ ╷ │
│ HTTP │ http://127.0.0.1:8020 │
│ Websocket │ ws://127.0.0.1:8020 │
│ ╵ │
╰───────────────────────────────────────╯
100%|████████████████████████████████████████████████| 9/9 [00:00<00:00, 44.15it/s]
100%|███████████████████████████████████████████████| 9/9 [00:00<00:00, 171.39it/s]

Scene type detection:

• Found 0/9 frames with significant sky presence (>20% sky pixels)
• Scene classified as: indoor, setting mask_sky to False
  (viser) Share URL requested!
  (viser) Generated share URL (expires in 24 hours, max 32 clients):
  https://tree-organic.share.viser.studio
  [13:17:42] Server 1 launched with URL demo.py:131
  https://tree-organic.share.viser.studio (pid: 37487) for
  session session_1741411059458
  (viser) Connection opened (0, 1 total), 124 persistent messages
  (viser) Connection opened (1, 2 total), 124 persistent messages

[jedyang97]

Is this result different from the hosted demo? Or are you just asking why it is bad? Without the images, I don't know what to expect.

[MYT999]

这个结果与托管演示不同吗？或者你只是想问为什么它不好？没有图像，我不知道会发生什么。
I want to know why the result is like this. Did I miss any steps? This is the image after running.

[jedyang97]

I see so these are the example scenes and you are asking why you are getting different results (much worse) than what the hosted demo outputs.

It is hard for me to tell exactly what is causing the discrepancy. But from my past experience, I have seen something like this when I installed the CroCo's cuROPE module. This is why I actually didn't include that step as in the DUST3R repo. But you mentioned that you followed the README step by step without skipping or adding steps, then I'm not too sure how to debug exactly the root cause.

Could you share your system's information? Like what GPU and what OS are you running?

[jedyang97]

Actually, looking at your logs, I don't see the warning line:

"Warning, cannot find cuda-compiled version of RoPE2D, using a slow pytorch version instead"

as in fast3r/croco/models/pos_embed.py. You probably have installed the CroCo's cuROPE. If so, please uninstall it by deleting the build folder. And make sure to follow the steps in the README as is.

[MYT999]

我明白了，这些是示例场景，而您要问为什么得到的结果与托管演示输出的结果不同（差得多）。

我很难确切地说出导致差异的原因。但根据我过去的经验，我在安装 CroCo 的 cuROPE 模块时看到过类似的东西。这就是为什么我实际上没有像在 DUST3R repo 中那样包含该步骤。但你提到你一步一步地遵循了 README，没有跳过或添加步骤，那么我不太确定如何准确地调试根本原因。

您能分享一下您的系统信息吗？例如您运行的是哪种 GPU 和哪种操作系统？

I'm currently in my first year of graduate school. I've been working on this for an entire day, but I still have no idea why it's not working. I have no choice but to ask for help here—I'm crying. I'm using Ubuntu 24.04 with an RTX 4060 Ti 16GB GPU.

[MYT999]

Okay, thank you! If I have any more questions, I'll come back to ask you.

[jedyang97]

I'm currently in my first year of graduate school. I've been working on this for an entire day, but I still have no idea why it's not working. I have no choice but to ask for help here—I'm crying.

I can totally understand the frustration. I've been there, too. Don't be too hard on yourself, and try to take it easy and enjoy research a bit! After all, we do research to advance science and fulfill our curiosity, don't we? ❤️ And of course, we will try to help you as much as we can!

[MYT999]

Okay, okay! Thank you so much! I'll try to stay relaxed while keeping up the hard work.