-
I am training a custom dataset using cosine-softmax, but all triplet loss, magnet loss, and cross entropy loss are being shown on tensorboard.
Which one should I be looking at when using cosine-so…
-
### Description
The current SOTA paper for image retrieval was published in 2020. It's called "Combination of Multiple Global Descriptors for Image Retrieval," and you can find it [here](https://arxi…
-
Hi,
is there any way to disable in-batch negatives during training in Sentence Transformers?
Thanks
@tomaarsen
-
Hi, thanks for sharing this amazing repository.
When using `aamsoftmax`, I am getting this error for all of the models.
```
Save path: exps/nist_old_bang_youtube_aamsoft_rawnet
Embedding size is…
-
I want to modify the file'/examples/triplet loss' and add some parameters into the trainer.
I have modified the file 'reid/trainers'
```
criterion1 = nn.CrossEntropyLoss().cuda()
criterion2 …
-
Training using google/mt5-base as the base model with fp16 and the triplet loss on all-nli data (following the [ example with trainer](https://www.sbert.net/docs/sentence_transformer/training_overview…
-
Any one know how to prepare data for triplet loss?
-
## T5-Generation Fine-tuning Task
### Data Preparation
- can only use specially-preprocessed: speaker-spec dataset, because the max input length for t5 is 512.
- preprocessing
1. `lib/create…
-
Hello,
When I tried to do the sh make.sh, I got the error:
g++: error: Hypothesis.o: No such file or directory
hough_voting_layer
make.sh: 67: make.sh: nvcc: not found
g++: error: roi_pooling_…
-
你好，有两个问题想要请教一下
1. 由于每张图片根据三个聚类结果有三个标签，代码里是根据第一个标签进行PK采样得到一个batch的图片数据，来进行triplet loss的计算。
但是，如何保证这个batch里的每一张图片在根据后两个聚类结果打上伪标签时，在这个batch中能找到正样本来保证triplet loss的正确计算呢？
2. 在selftraining.py文件中的compute_…