Skip to content

Commit cf972f6

Browse files
author
dbickson
committed
fixing
1 parent 19f39d3 commit cf972f6

File tree

1 file changed

+22
-3
lines changed

1 file changed

+22
-3
lines changed

RUN.md

Lines changed: 22 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -10,7 +10,8 @@
1010
7. [Performing vector search](#external)
1111
8. [Support for cloud storage](#s3)
1212
9. [Working with tar/zip files as input](#tar)
13-
10. [Debugging fastdup]
13+
10. [Working with video](#video)
14+
11. [Debugging fastdup](#debug)
1415

1516
## Detailed Python API documentation <a name="run"/>
1617
The main function of fastdup is `run`. It works by extracting short feature vectors from each image, clsutering the images together using a nearest neighbor model which computes similarities of pairs of images. Then a graph is formed to deduce the network structure of local similarities. The input/ outputs are described below in the section Input/Output.
@@ -380,6 +381,25 @@ For running on 50M images you will need an ubuntu machine with 32 cores and 256G
380381

381382

382383

384+
## Running video <a name="video"/>
385+
386+
fastdup supports video in mp4 and avi formats. For other formats please reach out. For running on video you need to install ffmpeg.
387+
388+
On Ubuntu
389+
```bash
390+
sudo apt install ffmpeg
391+
```
392+
393+
On Mac
394+
```bash
395+
brew install ffmpeg
396+
```
397+
398+
Note: on Mac 10.14 we encountered brew error, you can download statically compile ffmpeg [here](https://evermeet.cx/ffmpeg/).
399+
400+
Currently we extract frame 1 per sec, please reach out if you need other support. Our video tutorial is found here:
401+
- [🔥Analyzing video of the MEVA dataset - Google Colab](https://colab.research.google.com/github/visualdatabase/fastdup/blob/main/examples/fastdup_video.ipynb)
402+
383403

384404
## Debugging fastdup <a name="debug"/>
385405

@@ -388,5 +408,4 @@ To debug program execution the following is recommended
388408
- It is recommneded to debug in a python shell (and not in a Jupyter notebook)
389409
- Run with `verbose=1` to get additional traces
390410
- Run with `num_images=10` to run on a small subset of your data before running on the full dataset.
391-
- If the issue persist please join our slack channel, we would love to support!
392-
411+
- If the issue persist please join our [Slack Channel]("https://join.slack.com/t/visualdatabase/shared_invite/zt-19jaydbjn-lNDEDkgvSI1QwbTXSY6dlA")

0 commit comments

Comments
 (0)