vscode not showing images logged with log_image

[f-fuchs]

Hey,

when I use the vscode extension to compare experiments and show the plots of experiments the images I logged using log_image I just get the No Images to Compare.

The curious thing is, that when a single experiment is running and if I only select to show the plots of that experiment, the images do appear.

Could this be because I am caching the images and not adding them to git directly?

[shcheklein]

No, it should be working even if image plots are DVC-tracked. I've just checked with the example-repo - https://github.com/iterative/example-get-started-experiments - you can check eval stage and how images are handled there. VS Code extensions visualizes them fine.

Could you share some code, dvc.yaml how those are captured in your case? Vide / screenshot for the VS Code extension, and check also Show DVC Output command please in the VS Code.

[f-fuchs]

screenshot

dvc output

fatal: Could not read from remote repository.



Please make sure you have the correct access rights

and the repository exists.

[version: 1.2.27, 2025-01-28T07:19:54.204Z, pid: 1706] > git ls-files --others --exclude-standard - COMPLETED (45ms)
[version: 1.2.27, 2025-01-28T07:19:54.205Z, pid: 1708] > git ls-files --others --exclude-standard --directory --no-empty-directory - COMPLETED (39ms)
[version: 1.2.27, 2025-01-28T07:19:54.216Z, pid: 1710] > git rev-list --count HEAD - COMPLETED (29ms)
[version: 1.2.27, 2025-01-28T07:19:54.249Z, pid: 1709] > git log HEAD --pretty=format:%H%n%an%n%ar%nrefNames:%D%nmessage:%B -z -n 3 - COMPLETED (68ms)
[version: 1.2.27, 2025-01-28T07:19:54.260Z, pid: 1712] > /home/fuchsfa/wavelet-extraction-model/.venv/bin/python -m dvc exp show --rev b1f8fe080a2a97baceb18fee3de7cd6101f020a2 --rev 1523aadd71058de139f2aa4b162af3731253a0e8 --rev 130db65aebb6ac1ae3941048e24f0ad42f95f24e --json - INITIALIZED
[version: 1.2.27, 2025-01-28T07:19:56.674Z, pid: 1704] > /home/fuchsfa/wavelet-extraction-model/.venv/bin/python -m dvc data status --granular --unchanged --json - COMPLETED (2527ms)
[version: 1.2.27, 2025-01-28T07:19:58.031Z, pid: 1712] > /home/fuchsfa/wavelet-extraction-model/.venv/bin/python -m dvc exp show --rev b1f8fe080a2a97baceb18fee3de7cd6101f020a2 --rev 1523aadd71058de139f2aa4b162af3731253a0e8 --rev 130db65aebb6ac1ae3941048e24f0ad42f95f24e --json - COMPLETED (3780ms)

maybe the remote is the problem? but I only have a local remote and the experiments run on the same machine on which I try to visualize the images, therefore the images should be in the cache?

dvc.yaml

stages:
  train:
    cmd: python train.py
    deps:
    - train.py
    - src/
    params:
    - data
    - model
    - deterministic
    - training
    - plotting
    outs:
    - model.safetensors
    - dvclive/plots/images
  export_onnx:
    cmd: python export_onnx.py
    deps:
    - model.safetensors
    - export_onnx.py
    params:
    - onnx
    outs:
    - model.onnx
    - decoder.onnx
artifacts:
  model:
    path: model.onnx
    type: model
  decoder:
    path: decoder.onnx
    type: model
metrics:
- dvclive/metrics.json
plots:
- dvclive/plots/metrics:
    x: step

code

not sure what code u need. code is part of a bigger project. I am logging a matplotlib figure using
live.log_image(f"wavelet/{idx:03d}/{live.step}.png", fig)
and I create my Live instance with
with Live(monitor_system=False, report="html", cache_images=True) as live:

[f-fuchs]

okay I read this error wrong, I thought it meant the remote storage but it means the remote repository...

My git remote is behind 2fa, so I expect that to fail

[version: 1.2.27, 2025-01-29T07:29:16.949Z, pid: 179222] > git ls-remote origin refs/exps/* - FAILED with code 128 (62ms)
ssh: connect to host ssh2.itwm.fhg.de port 22: Connection refused


kex_exchange_identification: Connection closed by remote host


Connection closed by UNKNOWN port 65535


fatal: Could not read from remote repository.



Please make sure you have the correct access rights

and the repository exists.

[f-fuchs]

is it possible to disable this though? so that vscode is not spamming the server with requests that are bound to fail?

[shcheklein]

not sure if it's possible atm :( again, @mattseddon might remember better

[mattseddon]

doesn't look like this is possible

[f-fuchs]

okay 👍

[shcheklein]

plots:
- dvclive/plots/metrics:
    x: step

hmm, should it include images as well?

[f-fuchs]

I added it as an output to a stage. But I am not sure if this is correct 🤷
If I remember correctly I did not initially have it in the dvc.yaml at all and just followed what the prompts told me after running it.
but let me know what u think I should add there 👍

stages:
  train:
    cmd: python train.py
    deps:
    - train.py
    - src/
    params:
    - data
    - model
    - deterministic
    - training
    - plotting
    outs:
    - model.safetensors
    - dvclive/plots/images

[shcheklein]

do you have by chance a second dvc.yaml (dvclive/dvc.yaml or something) - could you try find -name dvc.yaml . or something

[f-fuchs]

nope, no second dvc.yaml just a .gigignore in dvclive/plots that ignores /images

[mattseddon]

@mattseddon could you remind please the logic behind it?

The three sections (data series, images and custom) have individual-size sliders. You can only change the width of images because we didn't want to mess with the aspect ratio, the max width of images is set to accommodate multiple images on the same row. You can open the image as a file by clicking on the image. You are then able to zoom that image using the normal VS Code behaviour: