# Discrepancy between Metadata and Web Visualizer

**URL:** <https://discuss.ego4d-data.org/t/discrepancy-between-metadata-and-web-visualizer/295>\
**Category:** Ego4D Feedback\
**Created:** [April 21, 2023, 8:20pm UTC](https://discuss.ego4d-data.org/t/discrepancy-between-metadata-and-web-visualizer/295 "2023-04-21T20:20:27Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![gweng](https://avatars.discourse-cdn.com/v4/letter/g/45deac/32.png) [@gweng](https://discuss.ego4d-data.org/u/gweng)\
**Post date:** [April 21, 2023, 8:20pm UTC](https://discuss.ego4d-data.org/t/discrepancy-between-metadata-and-web-visualizer/295/1 "2023-04-21T20:20:27Z")

</div>

Hi colleagues,

I found there might be some discrepancies between Metadata and Web Visualizer, at least for **gaze** data.

I read the latest Metadata `ego4d.json` from CLI, and downloaded the video ids filtered by `modalities include gaze`. Specifically, please check the screenshots attached.

 ![image](https://canada1.discourse-cdn.com/flex007/uploads/ego4d_data/original/1X/e02465780805b382b80c459b1cf917d3afd0a0e2.png)

However, the web visualizer told me [this video](https://visualize.ego4d-data.org/?q=%22modalities+include+gaze%22&s=%229b49246f-30e9-476f-ab8c-56a1bcb1d936%22) has gaze data.

It seems the Metadata is more reliable as I checked the gaze datasets, there is no corresponding `csv` file.

---

<div class="post-metadata">

**Author:** ![gweng](https://avatars.discourse-cdn.com/v4/letter/g/45deac/32.png) [@gweng](https://discuss.ego4d-data.org/u/gweng)\
**Post date:** [April 21, 2023, 9:31pm UTC](https://discuss.ego4d-data.org/t/discrepancy-between-metadata-and-web-visualizer/295/2 "2023-04-21T21:31:57Z")

</div>

I found that the `egp4d.json` aligns with the filtering results in the visualizer specified as `v2`, which is also annotated within the file. Here comes the following questions:

1. How can we download the `Metadata v1`?
2. How can we download `gaze v1` datasets? I tried  
`python -m ego4d.cli.cli --output_directory="*" --datasets imu gaze --version v1 --yes`  
The gaze data downloaded is the same as those in v2.

As I mentioned in another [post](https://discuss.ego4d-data.org/t/gaze-dataset-coverage-v1-vs-v2/292), there should be `2293` videos with gaze in `v1`.

---

<div class="post-metadata">

**Author:** ![gweng](https://avatars.discourse-cdn.com/v4/letter/g/45deac/32.png) [@gweng](https://discuss.ego4d-data.org/u/gweng)\
**Post date:** [April 21, 2023, 9:37pm UTC](https://discuss.ego4d-data.org/t/discrepancy-between-metadata-and-web-visualizer/295/3 "2023-04-21T21:37:21Z")

</div>

Is there a metafile that has complete information for all the latest available videos?

For example, the video `i` was released in `v1`, and `i` still exists in `v2`. The metafile should present `i` `has_gaze==True`.