search

donydchen / mvsplat

🌊 [ECCV'24 Oral] MVSplat: Efficient 3D Gaussian Splatting from Sparse Multi-View Images
https://donydchen.github.io/mvsplat
MIT License
790 stars 39 forks source link

training from scratch using the full RealEstate10k dataset? #71

Open emedinac opened 9 hours ago

emedinac commented 9 hours ago

Great work. Thanks for sharing!

It may be possible to train a new model from scratch using the full dataset from the RealEstate10K dataset. Did you try it? Some videos (YouTube links) were removed from the original dataset. I would like to know your opinion.

I have at least 572 sequences. Here is a short list:

0095ddd83beb3b8d
021d7121906d6cab
1ffd3b706d708774
214bae1626cba843
3c5bb5694853c36a
3f6c97f1ac96dada
5c55fe9bf93a3b33
5f1ad31daf0404f7
659a3f71c3383218
70f23e5d47f51d2b
7aad48affdc082a6
84d640e6f1b949a5
88820ffd5a224a0e
912a0872b546ecac
938f3602835af0df
978f6e10680c992d
b9963f1c210c909c
c60f84cee4f071a1
c6625644aad9430b
ddeb5bb5ca890490
e7a23e45161d8dce
ebe47ad7ed5b284d
eee5a58613ec6d4f
f7fd72838d272942
00ae21ab50209282
0a5e107e1961d01d
10eebcbb9021f437
23341c3a0b420e54
24ac24abf3057732
2bd43375196ce1a7
2cf32e2408107ea7
3172ad0d099430da
3916390b35258215
416f82fdbad68e21
44b2ab5292c06a7e
477a0a9f77c00480
7ba78a598d2d7abc
83a8899fb44ba5a6
9112263206128884
9852f2797fdf914d
987250887159290b
a96ba4ffa2c9cfcc
abd6e141bb826abe
b2ff375eff11279f
b863cfa14050f697
bb95f22eb2dd798e
c9a5a500538dfcea
d14239cd2f20bbfd
dd2d2046b390d101
e644440bff3bc30a
f37a1e10b2f4fdbf
02241c9f162966e3
06ffaa9ffc2eea95
27d1c320667bb5d3
3e02762a89de4c7d
4214dda1f9a61b1b
58b818a3eab9dc91
6c4e89d7af88b3c5
925440fbaa38398f
ab4b0f227cf2346f
b18dc9903ed6180d
b7ade000612a62c5
c6d255ece431d672
d376d04b2acd7ff0
dd62298bc00b4ba5
e7a59368d2f42369
037bdac76bdcd7c6
066ac822934d52d6
0761e0a3a4e25c53
11357e0934f26aed
20f1c3ade1608b4d
2ce75a0f430e2387
5cce543a7c11b825
67b8e88295424624
72e23a0c79923401
7974202a8aa9b7f1
80f5dfcec1498115
8fd34cca9299463e
a004746b1f88d95e
a1f682e249a9eb54
f0f881305e101463
04ed1812719e05f0
0b564b685315d8ff
164e33fbbcb5d223
2990bbc008d9fa82
3825ed66b7ea932e
39295d48af54153d
39f31c4461ede05f
484cfd0b6334c43a
5a6ebbfa6e3c0d03
8428647fab3bb509
94795b3aff58d9e3
975e3b9c0ee44348
990e2f1ddc581d87
ab8303a97b5834dc
af618ffabf190d3a
c6a2f7ebc0c4c497
ccf1bc0f2459d777
f2736f39d6213d22
05330a153a103386
0583c7a746238f79
0722819bdf5b2737
0b194567eff966f5
0e101fbd21daf79c
17d5dc2b775bdec2
2111cd087a82344b
2aab03e1aec0222c
3c91fa87850fd1a8
3d16e4b4719bc256
449ee1308d0710b5
4f25a98cd4b54467
5ffd56b01ed367fd
648f10c7d3f1b039
739dea1f7c339197
73cd755a2d69fe7b
7ae09aa6a1379a41
8f6b366f14d23d32
93c2bee2632822a5
980488bb1e65f544
a5a91020a894b511
a7c01c0765edb714
aadce44257d008ef
ad2169edb23a2753
bd0692bba8c57520
be6a659d766331e9
c1bb86f7ee52d4c3
ca98372626941bbb
cf3526b5c8f3cad1
d04e85080b5251c2
d29882026fc813ae
e44c67ca86fc2e99
0598fec76ecc7bd6
213286ef58a8c73d
353fffe2ffeaec91
60a79bac7c8a4e1d
6dd94ebca42cb8bf
844277dd18e25a54
a9228c6ca6619266
donydchen commented 4 hours ago

Hi @emedinac, thanks for your appreciation.

Yes, some YouTube videos might become unavailable for various reasons. A straightforward solution is to ignore these missing videos and train the model with those videos that are still available.

If you are interested in the preprocessed full datasets, drop me an email, and I will share them with you, including RE10K (~500G) and ACID (~160G). Cheers.

emedinac commented 3 hours ago

I sent you an email to: ...@ monash...

donydchen commented 3 hours ago

Shared @emedinac. Have fun training!

emedinac commented 3 hours ago

thanks!