fix(core): update build-map image metadata loading and imu propagator logging by junlinp · Pull Request #126 · UniflexAI/tinynav

junlinp · 2026-04-29T11:20:28Z

Summary

fix missing json import used by build_map_node metadata serialization
return lazy image loaders from TinyNavDB image accessor path
update imu_propagator_node logging behavior in callback path

Type

fix(core)

…data

…logging

github-actions · 2026-04-29T11:20:43Z

cloc	github.com/AlDanial/cloc v 2.04 T=0.14 s (305.9 files/s, 45731.9 lines/s)

Language	files	blank	comment	code
Python	13	620	259	3498
Markdown	4	155	6	426
C++	4	60	38	409
Bourne Shell	15	41	9	272
Dockerfile	1	30	29	191
JSON	2	8	0	58
TOML	1	4	0	58
make	1	18	15	55
CMake	1	4	0	17
--------	--------	--------	--------	--------
SUM:	42	940	356	4984

github-actions · 2026-04-29T11:28:17Z

cloc	github.com/AlDanial/cloc v 2.04 T=0.17 s (245.0 files/s, 36547.8 lines/s)

Language	files	blank	comment	code
Python	13	620	259	3483
Markdown	4	155	6	426
C++	4	60	38	409
Bourne Shell	15	41	9	272
Dockerfile	1	30	29	191
JSON	2	8	0	58
TOML	1	4	0	58
make	1	18	15	55
CMake	1	4	0	17
--------	--------	--------	--------	--------
SUM:	42	940	356	4969

github-actions · 2026-04-30T02:11:09Z

cloc	github.com/AlDanial/cloc v 2.04 T=0.13 s (324.1 files/s, 48657.9 lines/s)

Language	files	blank	comment	code
Python	13	626	262	3514
Markdown	4	155	6	426
C++	4	60	38	409
Bourne Shell	15	41	9	272
Dockerfile	1	30	29	191
JSON	2	8	0	58
TOML	1	4	0	58
make	1	18	15	55
CMake	1	4	0	17
--------	--------	--------	--------	--------
SUM:	42	946	359	5000

dvorak0 · 2026-04-30T02:29:51Z

https://github.com/dmlc/decord useful?

junlinp · 2026-04-30T02:52:47Z

https://github.com/dmlc/decord useful?

Decord provides efficient random-access capabilities and integrates well with hardware decoders. We also need to generate timestamp-to-frame-index metadata using PyAV. The build-map node only saves the images and does not read them afterward. Decord is therefore a better choice when these images are used to train a model.

github-actions · 2026-04-30T03:07:04Z

cloc	github.com/AlDanial/cloc v 2.04 T=0.13 s (327.2 files/s, 49118.1 lines/s)

Language	files	blank	comment	code
Python	13	626	262	3514
Markdown	4	155	6	426
C++	4	60	38	409
Bourne Shell	15	41	9	272
Dockerfile	1	30	29	191
JSON	2	8	0	58
TOML	1	4	0	58
make	1	18	15	55
CMake	1	4	0	17
--------	--------	--------	--------	--------
SUM:	42	946	359	5000

dvorak0 · 2026-04-30T03:20:26Z

@junlinp It's too messy to have so much of code doing "writing". Can we have a better one?

Like a in-house package of VideoDB with interface of:

open("infra1") gives a folder infra1
write(timestamp, Image)
read(timestamp, Image)
iter()

So we could hide all timestamp2index inside the video_db.py

It would be great if we do statistics (avg encode/decode time per-frame) inside video_db.py, as I'm concerned about the CPU consumption of encode/decode.

We end up with two kind of DB in the folder: 1. generic key-value shelve DB; 2. key-Image Video DB.

❯ ls -l
.rw-r--r--  136 root 20 Apr 16:22 baseline.npy
.rw-r--r-- 4.0G root 20 Apr 16:22 depths.db
.rw-r--r--  39k root 20 Apr 16:22 edges.npy
.rw-r--r--  11M root 20 Apr 16:22 embeddings.db
.rw-r--r-- 1.3G root 20 Apr 16:22 features.db
.rw-r--r-- 1.0G root 20 Apr 16:22 infra1_images.db
.rw-r--r--  200 root 20 Apr 16:22 intrinsics.npy
.rw-r--r-- 697k root 20 Apr 16:22 mapping_continuous_odom.npy
.rw-r--r-- 4.1k root 20 Apr 16:22 occupancy_2d_image.png
.rw-r--r-- 2.8M root 20 Apr 16:22 occupancy_grid.npy
.rw-r--r--  144 root 20 Apr 16:22 occupancy_meta.npy
.rw-r--r--  451 root 20 Apr 17:44 pois.json
.rw-r--r-- 467k root 20 Apr 16:22 poses.npy
.rw-r--r--  200 root 20 Apr 16:22 rgb_camera_intrinsics.npy
.rw-r--r-- 3.0G root 20 Apr 16:22 rgb_images.db
.rw-r--r--  11M root 20 Apr 16:22 sdf_map.npy
.rw-r--r--  278 root 20 Apr 16:22 T_rgb_to_infra1.npy

to

❯ ls -l
.rw-r--r--  136 root 20 Apr 16:22 baseline.npy
.rw-r--r-- 4.0G root 20 Apr 16:22 depths.db
.rw-r--r--  39k root 20 Apr 16:22 edges.npy
.rw-r--r--  11M root 20 Apr 16:22 embeddings.db
.rw-r--r-- 1.3G root 20 Apr 16:22 features.db
.rw-r--r-- 1.0G root 20 Apr 16:22 infra1_images_db/
.rw-r--r--  200 root 20 Apr 16:22 intrinsics.npy
.rw-r--r-- 697k root 20 Apr 16:22 mapping_continuous_odom.npy
.rw-r--r-- 4.1k root 20 Apr 16:22 occupancy_2d_image.png
.rw-r--r-- 2.8M root 20 Apr 16:22 occupancy_grid.npy
.rw-r--r--  144 root 20 Apr 16:22 occupancy_meta.npy
.rw-r--r--  451 root 20 Apr 17:44 pois.json
.rw-r--r-- 467k root 20 Apr 16:22 poses.npy
.rw-r--r--  200 root 20 Apr 16:22 rgb_camera_intrinsics.npy
.rw-r--r-- 3.0G root 20 Apr 16:22 rgb_images_db/
.rw-r--r--  11M root 20 Apr 16:22 sdf_map.npy
.rw-r--r--  278 root 20 Apr 16:22 T_rgb_to_infra1.npy

infra1_images.db -> infra1_images_db/

github-actions · 2026-04-30T04:15:54Z

cloc	github.com/AlDanial/cloc v 2.04 T=0.17 s (244.7 files/s, 36856.9 lines/s)

Language	files	blank	comment	code
Python	13	625	262	3537
Markdown	4	155	6	426
C++	4	60	38	409
Bourne Shell	15	41	9	272
Dockerfile	1	30	29	191
JSON	2	8	0	58
TOML	1	4	0	58
make	1	18	15	55
CMake	1	4	0	17
--------	--------	--------	--------	--------
SUM:	42	945	359	5023

github-actions · 2026-04-30T04:18:36Z

cloc	github.com/AlDanial/cloc v 2.04 T=0.17 s (249.4 files/s, 37575.2 lines/s)

Language	files	blank	comment	code
Python	13	625	262	3538
Markdown	4	155	6	426
C++	4	60	38	409
Bourne Shell	15	41	9	272
Dockerfile	1	30	29	191
JSON	2	8	0	58
TOML	1	4	0	58
make	1	18	15	55
CMake	1	4	0	17
--------	--------	--------	--------	--------
SUM:	42	945	359	5024

junlinp · 2026-04-30T04:21:49Z

The encoding time for the RGB task is 20 ms in pc

github-actions · 2026-04-30T04:25:01Z

cloc	github.com/AlDanial/cloc v 2.04 T=0.17 s (249.6 files/s, 37602.8 lines/s)

Language	files	blank	comment	code
Python	13	625	262	3538
Markdown	4	155	6	426
C++	4	60	38	409
Bourne Shell	15	41	9	272
Dockerfile	1	30	29	191
JSON	2	8	0	58
TOML	1	4	0	58
make	1	18	15	55
CMake	1	4	0	17
--------	--------	--------	--------	--------
SUM:	42	945	359	5024

dvorak0 · 2026-04-30T06:04:57Z

The encoding time for the RGB task is 20 ms in pc

@junlinp For sure my concern is around the jetson. And would you mind moving VideoDB into tool, so we could make the core code really "core".

junlinp · 2026-04-30T06:10:36Z

The encoding time for the RGB task is 20 ms in pc

@junlinp For sure my concern is around the jetson. And would you mind moving VideoDB into tool, so we could make the core code really "core".

I see. I’ve modified the conversion tool to benchmark read performance and integrated VideoDB into the tool to make the core logic cleaner.

github-actions · 2026-04-30T07:12:23Z

cloc	github.com/AlDanial/cloc v 2.04 T=0.13 s (320.1 files/s, 47209.4 lines/s)

Language	files	blank	comment	code
Python	13	615	262	3414
Markdown	4	155	6	426
C++	4	60	38	409
Bourne Shell	15	41	9	272
Dockerfile	1	30	29	191
TOML	1	4	0	59
JSON	2	8	0	58
make	1	18	15	55
CMake	1	4	0	17
--------	--------	--------	--------	--------
SUM:	42	935	359	4901

dvorak0 · 2026-04-30T07:13:20Z

@junlinp looks good. let me know if it's ready to merge.

junlinp · 2026-04-30T10:14:00Z

img_v3_02118_884216f3-fcd8-4db0-a61a-ee15cec968hu

simple test on jetson,avg read :0.5ms, avg write: 8ms. I think this is acceptable.

github-actions · 2026-04-30T10:16:16Z

cloc	github.com/AlDanial/cloc v 2.04 T=0.17 s (245.4 files/s, 36259.4 lines/s)

Language	files	blank	comment	code
Python	13	615	262	3414
Markdown	4	155	6	426
C++	4	60	38	409
Bourne Shell	15	41	9	272
Dockerfile	1	31	30	200
JSON	2	8	0	58
TOML	1	4	0	58
make	1	18	15	55
CMake	1	4	0	17
--------	--------	--------	--------	--------
SUM:	42	936	360	4909

junlinp · 2026-04-30T10:18:19Z

@dvorak0 This PR is ready to be merged, and the Docker image needs to be updated.

dvorak0 · 2026-04-30T10:43:46Z

@junlinp shall we just fail fast when decord is missing. Try catch wont save us.

github-actions · 2026-04-30T10:49:47Z

cloc	github.com/AlDanial/cloc v 2.04 T=0.19 s (225.4 files/s, 33296.0 lines/s)

Language	files	blank	comment	code
Python	13	615	262	3414
Markdown	4	155	6	426
C++	4	60	38	409
Bourne Shell	15	41	9	272
Dockerfile	1	31	30	200
JSON	2	8	0	58
TOML	1	4	0	58
make	1	18	15	55
CMake	1	4	0	17
--------	--------	--------	--------	--------
SUM:	42	936	360	4909

junlinp · 2026-04-30T10:53:59Z

@junlinp shall we just fail fast when decord is missing. Try catch wont save us.

yep.the AI tends to over-defensive programming.

dvorak0 · 2026-04-30T11:05:45Z

@junlinp shall we just fail fast when decord is missing. Try catch wont save us.

yep.the AI tends to over-defensive programming.

@junlinp you can merge this as you want. Note CI fails now.

junlinp added 3 commits April 29, 2026 11:48

feat(build_map): store image streams in mp4 with timestamp index meta…

6ca6462

…data

fix(build_map): avoid unnecessary mp4 image decode in db reads

455292d

fix(core): update build map image metadata import and imu propagator …

701fc6c

…logging

chore(imu): remove callback imu logging counter

984731e

chore(lock): update uv lockfile

241865b

feat(build_map): set mp4 writer fps to 30

9a2e922

junlinp marked this pull request as ready for review April 30, 2026 03:09

junlinp requested a review from dvorak0 April 30, 2026 03:09

refactor(build_map): split video storage into directory-backed VideoDB

67116ad

refactor(build_map): enforce consistent image ndim in VideoDB writes

4810c2c

fix(build_map): resolve ruff import shadowing in transform publish

c31c57e

junlinp added 2 commits April 30, 2026 14:58

feat(tool): add VideoDB with near-random-access frame decode

4dd9996

refactor(build_map): extract VideoDB and switch nerf export to video API

6fb5c6d

dvorak0 approved these changes Apr 30, 2026

View reviewed changes

build(docker): build decord from source and remove pip dependency

58104af

build(docker): update decord source build steps

692f78e

refactor(tool): use VideoDB API for nerf export and decord-backed reads

d141197

Conversation

junlinp commented Apr 29, 2026

Summary

Type

Uh oh!

github-actions Bot commented Apr 29, 2026

Uh oh!

github-actions Bot commented Apr 29, 2026

Uh oh!

github-actions Bot commented Apr 30, 2026

Uh oh!

dvorak0 commented Apr 30, 2026

Uh oh!

junlinp commented Apr 30, 2026

Uh oh!

github-actions Bot commented Apr 30, 2026

Uh oh!

dvorak0 commented Apr 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Apr 30, 2026

Uh oh!

github-actions Bot commented Apr 30, 2026

Uh oh!

junlinp commented Apr 30, 2026

Uh oh!

github-actions Bot commented Apr 30, 2026

Uh oh!

dvorak0 commented Apr 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

junlinp commented Apr 30, 2026

Uh oh!

github-actions Bot commented Apr 30, 2026

Uh oh!

dvorak0 commented Apr 30, 2026

Uh oh!

junlinp commented Apr 30, 2026

Uh oh!

github-actions Bot commented Apr 30, 2026

Uh oh!

junlinp commented Apr 30, 2026

Uh oh!

dvorak0 commented Apr 30, 2026

Uh oh!

github-actions Bot commented Apr 30, 2026

Uh oh!

junlinp commented Apr 30, 2026

Uh oh!

dvorak0 commented Apr 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

dvorak0 commented Apr 30, 2026 •

edited

Loading

dvorak0 commented Apr 30, 2026 •

edited

Loading

dvorak0 commented Apr 30, 2026 •

edited

Loading