Bill Psomas commited on
Commit
ecc3317
1 Parent(s): 7dc9032

update readme

Browse files
Files changed (2) hide show
  1. README.md +10 -0
  2. resnet50_dino_official.yaml +2 -2
README.md CHANGED
@@ -23,6 +23,16 @@ ResNet-50 official model trained on ImageNet-1k for 100 epochs. Self-supervision
23
  SimPool is a simple attention-based pooling method at the end of network, released in this [repository](https://github.com/billpsomas/simpool/).
24
  Disclaimer: This model card is written by the author of SimPool, i.e. [Bill Psomas](http://users.ntua.gr/psomasbill/).
25
 
 
 
 
 
 
 
 
 
 
 
26
  ## BibTeX entry and citation info
27
 
28
  ```
 
23
  SimPool is a simple attention-based pooling method at the end of network, released in this [repository](https://github.com/billpsomas/simpool/).
24
  Disclaimer: This model card is written by the author of SimPool, i.e. [Bill Psomas](http://users.ntua.gr/psomasbill/).
25
 
26
+ ## Evaluation with k-NN
27
+
28
+ | k | top1 | top5 |
29
+ | ------- | ------- | ------- |
30
+ | 10 | 61.84 | 80.35 |
31
+ | 20 | 62.174 | 82.75 |
32
+ | 100 | 60.088 | 84.216 |
33
+ | 200 | 58.544 | 83.834 |
34
+
35
+
36
  ## BibTeX entry and citation info
37
 
38
  ```
resnet50_dino_official.yaml CHANGED
@@ -1,12 +1,12 @@
1
  arch: resnet50
2
  backend: nccl
3
- batch_size_per_gpu: 90
4
  clip_grad: 0.0
5
  data_path: /path/to/imagenet/
6
  dist_url: env://
7
  drop_path_rate: 0.1
8
  epochs: 100
9
- eval_every: 5
10
  freeze_last_layer: 1
11
  global_crops_scale:
12
  - 0.14
 
1
  arch: resnet50
2
  backend: nccl
3
+ batch_size_per_gpu: 80
4
  clip_grad: 0.0
5
  data_path: /path/to/imagenet/
6
  dist_url: env://
7
  drop_path_rate: 0.1
8
  epochs: 100
9
+ eval_every: 10
10
  freeze_last_layer: 1
11
  global_crops_scale:
12
  - 0.14