Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -41,4 +41,22 @@ The LSDBench dataset is designed to evaluate the sampling efficiency of long-vid
|
|
41 |
|
42 |
## Evaluation on LSDBench
|
43 |
|
44 |
-
Please see our [github repo](https://github.com/dvlab-research/LSDBench) for detailed evaluation guide.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
|
42 |
## Evaluation on LSDBench
|
43 |
|
44 |
+
Please see our [github repo](https://github.com/dvlab-research/LSDBench) for detailed evaluation guide.
|
45 |
+
|
46 |
+
<div align=left>
|
47 |
+
<img width="45%" src="https://raw.githubusercontent.com/dvlab-research/LSDBench/main/assets/main_table.png"/>
|
48 |
+
</div>
|
49 |
+
|
50 |
+
_**Performance comparison of different models and sampling
|
51 |
+
strategies.** We present three testing settings in total: Only
|
52 |
+
Text, Oracle, and Full Video. In the Only Text setting, the model is
|
53 |
+
provided with no visual information whatsoever. The Oracle setting
|
54 |
+
involves using the annotated target segment as the video input,
|
55 |
+
while the Full Video setting provides the complete long video as
|
56 |
+
input. The ”Sampling” column lists the sampling strategies used:
|
57 |
+
FPS represents sampling at fixed time intervals, fixed denotes uniform
|
58 |
+
sampling with a fixed number of frames, and 2stage refers
|
59 |
+
to the method we propose. Under each sampling strategy, the average
|
60 |
+
number of sampled frames during evaluation on the LSDBench
|
61 |
+
dataset is reported in the ”Frames” column, along with the
|
62 |
+
corresponding sampling density (SD)._
|