forked from sermanet/OverFeat
-
Notifications
You must be signed in to change notification settings - Fork 0
/
CHANGELOG
46 lines (37 loc) · 1.69 KB
/
CHANGELOG
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
version 3:
- Fix Makefile for Mac OS.
- The torch wrapper now takes inputs images where the pixel values go from 0 to 1 (it used to be from 0 to 255) to harmonize with the torch image library.
- The option -f now returns the output of layer 18 (it used to be layer 16), so that the output should have size 4096x1x1 if the input has size 3x231x231.
- Add python wrapper.
- Add Mac OS binaries.
version 3.1:
- Fix overfeat script so that the correct layer is produced for the large network.
version 3.2:
- Fix bug in python wrapper that caused some images to be incorrectly loaded.
- Better python wrapper sample script.
version 3.3:
- Fix overfeat script that resized the images even when extracting the features.
- Changed the default output layer of the overfeat script when using feature extraction.
- Separate files from weights and add download_weights.py script to download the weights.
version 3.4:
- Added binaries for linux 32 bits.
- Fixed wrong output layer in overseat with option -f.
version 3.5:
- Fix LD_LIBRARY_PATH in overfeat script.
- Add git repository.
version 3.6:
- Fix ppm images reading that didn't support comments in the ppm image.
- Fix wrong paths in overfeat_batch.
- Fix layer numbers in README.
version 4.0:
- Released binaries for GPU (only for linux 64bits for now).
- Removed dependency to torch by adding local copy of lib TH.
- Replaced dynamic library libTH.so by static library libTH.a.
- Fix bug in webcam script.
- Fixed ambiguous dependency to openblas : now, linux uses openblas and macos uses Accelerate.
- Mac OS should correctly use OpenMP now.
version 4.1:
- Fix python API
- Add missing file for cuda C++ API
version 4.2:
- Fix GPU feature extractor