Add UZH-FPV drone racing dataset (UZHFPV) by vlordier · Pull Request #331 · neuromorphs/tonic

vlordier · 2026-03-25T13:30:27Z

Co-authored-by: vlordier 5443125+vlordier@users.noreply.github.com
Agent-Logs-Url: https://github.com/vlordier/tonic/sessions/33daba03-974f-4ca9-a7f3-967a32a6a818

Co-authored-by: vlordier <5443125+vlordier@users.noreply.github.com> Agent-Logs-Url: https://github.com/vlordier/tonic/sessions/33daba03-974f-4ca9-a7f3-967a32a6a818

codecov-commenter · 2026-03-25T13:36:02Z

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 41.66667% with 42 lines in your changes missing coverage. Please review.
✅ Project coverage is 74.21%. Comparing base (8f95aec) to head (6bdb240).
⚠️ Report is 5 commits behind head on develop.

Files with missing lines	Patch %	Lines
tonic/datasets/uzh_fpv.py	40.84%	42 Missing ⚠️
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop     #331      +/-   ##
===========================================
- Coverage    76.50%   74.21%   -2.29%     
===========================================
  Files           48       50       +2     
  Lines         2992     3126     +134     
===========================================
+ Hits          2289     2320      +31     
- Misses         703      806     +103

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

Copilot

Pull request overview

Adds support for the UZH-FPV drone racing dataset to Tonic, including a new Dataset implementation, package exports, documentation listing, and basic tests.

Changes:

Introduce tonic.datasets.UZHFPV to download/read UZH-FPV ROS bag recordings (events/IMU/images + optional GT).
Export UZHFPV from tonic.datasets and include it in the datasets docs autosummary.
Add unit tests for recording selection and validation.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 4 comments.

File	Description
`tonic/datasets/uzh_fpv.py`	New dataset implementation for UZH-FPV, including download, existence checks, and ROS bag parsing.
`tonic/datasets/__init__.py`	Exposes `UZHFPV` in the datasets package namespace and `__all__`.
`test/test_uzh_fpv.py`	Adds tests for recording selection (`single`, `multiple`, `all`) and validation behavior.
`docs/datasets.rst`	Lists `UZHFPV` under pose estimation / odometry / SLAM datasets.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-03-25T13:37:35Z

+        # Normalize timestamps to start at 0 and convert from seconds to microseconds
+        events = topics["/dvs/events"]
+        events["ts"] -= events["ts"][0]
+        events["ts"] *= 1e6
+        events = make_structured_array(
+            events["ts"], events["x"], events["y"], events["pol"], dtype=self.dtype
+        )
+
+        # Normalize and convert IMU timestamps to microseconds
+        imu = topics["/dvs/imu"]
+        imu["ts"] = ((imu["ts"] - imu["ts"][0]) * 1e6).astype(int)
+
+        # Normalize and convert image timestamps to microseconds, stack frames
+        images = topics["/dvs/image_raw"]
+        images["frames"] = np.stack(images["frames"])
+        images["ts"] = ((images["ts"] - images["ts"][0]) * 1e6).astype(int)
+
+        data = events, imu, images
+
+        if has_gt and "/groundtruth/odometry" in topics:
+            # Normalize and convert ground truth timestamps to microseconds
+            target = topics["/groundtruth/odometry"]
+            target["ts"] = ((target["ts"] - target["ts"][0]) * 1e6).astype(int)


Ground-truth timestamps are normalized against the first GT sample, which can desynchronize GT from events/IMU/images if those streams use different t0. Normalize GT timestamps using the same shared reference time as the rest of the modalities so consumers can directly compare timelines.

Suggested change

# Normalize timestamps to start at 0 and convert from seconds to microseconds

events = topics["/dvs/events"]

events["ts"] -= events["ts"][0]

events["ts"] *= 1e6

events = make_structured_array(

events["ts"], events["x"], events["y"], events["pol"], dtype=self.dtype

)

# Normalize and convert IMU timestamps to microseconds

imu = topics["/dvs/imu"]

imu["ts"] = ((imu["ts"] - imu["ts"][0]) * 1e6).astype(int)

# Normalize and convert image timestamps to microseconds, stack frames

images = topics["/dvs/image_raw"]

images["frames"] = np.stack(images["frames"])

images["ts"] = ((images["ts"] - images["ts"][0]) * 1e6).astype(int)

data = events, imu, images

if has_gt and "/groundtruth/odometry" in topics:

# Normalize and convert ground truth timestamps to microseconds

target = topics["/groundtruth/odometry"]

target["ts"] = ((target["ts"] - target["ts"][0]) * 1e6).astype(int)

# Use a shared reference time (t0) based on the first event timestamp

base_ts = topics["/dvs/events"]["ts"][0]

# Normalize event timestamps to start at 0 (shared t0) and convert from seconds to microseconds

events = topics["/dvs/events"]

events["ts"] = (events["ts"] - base_ts) * 1e6

events = make_structured_array(

events["ts"], events["x"], events["y"], events["pol"], dtype=self.dtype

)

# Normalize and convert IMU timestamps to microseconds using the shared reference time

imu = topics["/dvs/imu"]

imu["ts"] = ((imu["ts"] - base_ts) * 1e6).astype(int)

# Normalize and convert image timestamps to microseconds using the shared reference time, stack frames

images = topics["/dvs/image_raw"]

images["frames"] = np.stack(images["frames"])

images["ts"] = ((images["ts"] - base_ts) * 1e6).astype(int)

data = events, imu, images

if has_gt and "/groundtruth/odometry" in topics:

# Normalize and convert ground truth timestamps to microseconds using the shared reference time

target = topics["/groundtruth/odometry"]

target["ts"] = ((target["ts"] - base_ts) * 1e6).astype(int)

Copilot · 2026-03-25T13:37:35Z

+def test_single_recording():
+    with patch.object(tonic.datasets.UZHFPV, "_check_exists", return_value=True):
+        dataset = tonic.datasets.UZHFPV(
+            save_to="data",
+            recording="indoor_forward_3",
+        )
+    assert len(dataset) == 1


Current tests only validate recording selection/validation and don’t exercise getitem behavior (ROS bag parsing, timestamp conversion/normalization, transform hooks, and GT presence/absence). Add a unit test that patches importRosbag to return a minimal synthetic topics dict and asserts the returned events dtype/ordering, timestamp units, and that data/target transforms are applied as expected.

Copilot · 2026-03-25T13:37:36Z

+        # Normalize timestamps to start at 0 and convert from seconds to microseconds
+        events = topics["/dvs/events"]
+        events["ts"] -= events["ts"][0]
+        events["ts"] *= 1e6
+        events = make_structured_array(
+            events["ts"], events["x"], events["y"], events["pol"], dtype=self.dtype
+        )
+
+        # Normalize and convert IMU timestamps to microseconds
+        imu = topics["/dvs/imu"]
+        imu["ts"] = ((imu["ts"] - imu["ts"][0]) * 1e6).astype(int)
+
+        # Normalize and convert image timestamps to microseconds, stack frames
+        images = topics["/dvs/image_raw"]
+        images["frames"] = np.stack(images["frames"])
+        images["ts"] = ((images["ts"] - images["ts"][0]) * 1e6).astype(int)
+
+        data = events, imu, images
+
+        if has_gt and "/groundtruth/odometry" in topics:
+            # Normalize and convert ground truth timestamps to microseconds
+            target = topics["/groundtruth/odometry"]
+            target["ts"] = ((target["ts"] - target["ts"][0]) * 1e6).astype(int)


Timestamp normalization uses each stream’s own first timestamp as t0 (events, IMU, images). This breaks cross-sensor alignment (e.g., IMU/image timestamps won’t be comparable to event timestamps). Consider computing a single reference time (e.g., the earliest timestamp across available topics, or the events’ first timestamp) and subtracting that from all modalities before converting to microseconds.

Suggested change

# Normalize timestamps to start at 0 and convert from seconds to microseconds

events = topics["/dvs/events"]

events["ts"] -= events["ts"][0]

events["ts"] *= 1e6

events = make_structured_array(

events["ts"], events["x"], events["y"], events["pol"], dtype=self.dtype

)

# Normalize and convert IMU timestamps to microseconds

imu = topics["/dvs/imu"]

imu["ts"] = ((imu["ts"] - imu["ts"][0]) * 1e6).astype(int)

# Normalize and convert image timestamps to microseconds, stack frames

images = topics["/dvs/image_raw"]

images["frames"] = np.stack(images["frames"])

images["ts"] = ((images["ts"] - images["ts"][0]) * 1e6).astype(int)

data = events, imu, images

if has_gt and "/groundtruth/odometry" in topics:

# Normalize and convert ground truth timestamps to microseconds

target = topics["/groundtruth/odometry"]

target["ts"] = ((target["ts"] - target["ts"][0]) * 1e6).astype(int)

# Extract raw modalities

events = topics["/dvs/events"]

imu = topics["/dvs/imu"]

images = topics["/dvs/image_raw"]

# Compute a common reference time (earliest timestamp across available streams)

t0_candidates = [events["ts"][0], imu["ts"][0], images["ts"][0]]

gt_topic = None

if has_gt and "/groundtruth/odometry" in topics:

gt_topic = topics["/groundtruth/odometry"]

t0_candidates.append(gt_topic["ts"][0])

t0 = min(t0_candidates)

# Normalize event timestamps to common t0 and convert from seconds to microseconds

events_ts = ((events["ts"] - t0) * 1e6).astype(int)

events = make_structured_array(

events_ts, events["x"], events["y"], events["pol"], dtype=self.dtype

)

# Normalize and convert IMU timestamps to microseconds using common t0

imu["ts"] = ((imu["ts"] - t0) * 1e6).astype(int)

# Normalize and convert image timestamps to microseconds using common t0, stack frames

images["frames"] = np.stack(images["frames"])

images["ts"] = ((images["ts"] - t0) * 1e6).astype(int)

data = events, imu, images

if gt_topic is not None:

# Normalize and convert ground truth timestamps to microseconds using common t0

target = gt_topic

target["ts"] = ((target["ts"] - t0) * 1e6).astype(int)

Copilot · 2026-03-25T13:37:36Z

+        events["ts"] -= events["ts"][0]
+        events["ts"] *= 1e6
+        events = make_structured_array(
+            events["ts"], events["x"], events["y"], events["pol"], dtype=self.dtype
+        )
+
+        # Normalize and convert IMU timestamps to microseconds
+        imu = topics["/dvs/imu"]
+        imu["ts"] = ((imu["ts"] - imu["ts"][0]) * 1e6).astype(int)
+
+        # Normalize and convert image timestamps to microseconds, stack frames
+        images = topics["/dvs/image_raw"]
+        images["frames"] = np.stack(images["frames"])
+        images["ts"] = ((images["ts"] - images["ts"][0]) * 1e6).astype(int)
+
+        data = events, imu, images
+
+        if has_gt and "/groundtruth/odometry" in topics:
+            # Normalize and convert ground truth timestamps to microseconds
+            target = topics["/groundtruth/odometry"]
+            target["ts"] = ((target["ts"] - target["ts"][0]) * 1e6).astype(int)


IMU timestamps are normalized relative to the first IMU sample rather than the same time origin as events/images. For synchronized sensor fusion use-cases, normalize IMU (and image) timestamps against the same reference time used for events so timelines remain consistent.

Suggested change

events["ts"] -= events["ts"][0]

events["ts"] *= 1e6

events = make_structured_array(

events["ts"], events["x"], events["y"], events["pol"], dtype=self.dtype

)

# Normalize and convert IMU timestamps to microseconds

imu = topics["/dvs/imu"]

imu["ts"] = ((imu["ts"] - imu["ts"][0]) * 1e6).astype(int)

# Normalize and convert image timestamps to microseconds, stack frames

images = topics["/dvs/image_raw"]

images["frames"] = np.stack(images["frames"])

images["ts"] = ((images["ts"] - images["ts"][0]) * 1e6).astype(int)

data = events, imu, images

if has_gt and "/groundtruth/odometry" in topics:

# Normalize and convert ground truth timestamps to microseconds

target = topics["/groundtruth/odometry"]

target["ts"] = ((target["ts"] - target["ts"][0]) * 1e6).astype(int)

reference_time = events["ts"][0]

events["ts"] = (events["ts"] - reference_time) * 1e6

events = make_structured_array(

events["ts"], events["x"], events["y"], events["pol"], dtype=self.dtype

)

# Normalize and convert IMU timestamps to microseconds using the same reference time as events

imu = topics["/dvs/imu"]

imu["ts"] = ((imu["ts"] - reference_time) * 1e6).astype(int)

# Normalize and convert image timestamps to microseconds, stack frames, using the same reference time

images = topics["/dvs/image_raw"]

images["frames"] = np.stack(images["frames"])

images["ts"] = ((images["ts"] - reference_time) * 1e6).astype(int)

data = events, imu, images

if has_gt and "/groundtruth/odometry" in topics:

# Normalize and convert ground truth timestamps to microseconds using the same reference time

target = topics["/groundtruth/odometry"]

target["ts"] = ((target["ts"] - reference_time) * 1e6).astype(int)

vlordier · 2026-03-25T13:40:59Z

@copilot open a new pull request to apply changes based on the comments in this thread

vlordier · 2026-03-25T13:41:49Z

@copilot fix coverage : ❌ Patch coverage is 41.66667% with 42 lines in your changes missing coverage. Please review.

vlordier · 2026-03-25T16:28:53Z

@copilot open a new pull request to apply changes based on the comments in this thread

Add UZH-FPV drone racing dataset (UZHFPV)

6bdb240

Co-authored-by: vlordier <5443125+vlordier@users.noreply.github.com> Agent-Logs-Url: https://github.com/vlordier/tonic/sessions/33daba03-974f-4ca9-a7f3-967a32a6a818

Copilot AI review requested due to automatic review settings March 25, 2026 13:30

Copilot started reviewing on behalf of vlordier March 25, 2026 13:30 View session

Copilot AI reviewed Mar 25, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add UZH-FPV drone racing dataset (UZHFPV)#331

Add UZH-FPV drone racing dataset (UZHFPV)#331
vlordier wants to merge 1 commit intoneuromorphs:developfrom
vlordier:copilot/add-datasets-from-fpv-ifi-uzh-ch

vlordier commented Mar 25, 2026

Uh oh!

codecov-commenter commented Mar 25, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Mar 25, 2026

Uh oh!

Copilot AI Mar 25, 2026

Uh oh!

Copilot AI Mar 25, 2026

Uh oh!

Copilot AI Mar 25, 2026

Uh oh!

vlordier commented Mar 25, 2026

Uh oh!

vlordier commented Mar 25, 2026

Uh oh!

vlordier commented Mar 25, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

vlordier commented Mar 25, 2026

Uh oh!

codecov-commenter commented Mar 25, 2026

Codecov Report

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Mar 25, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 25, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 25, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 25, 2026

Choose a reason for hiding this comment

Uh oh!

vlordier commented Mar 25, 2026

Uh oh!

vlordier commented Mar 25, 2026

Uh oh!

vlordier commented Mar 25, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants