Skip to content

Commit

Permalink
small fixes
Browse files Browse the repository at this point in the history
  • Loading branch information
JSabadin committed Jan 16, 2025
1 parent 04a6ab6 commit 2ceacc5
Show file tree
Hide file tree
Showing 2 changed files with 7 additions and 5 deletions.
2 changes: 2 additions & 0 deletions luxonis_train/config/predefined_models/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -162,6 +162,8 @@ See an example configuration file using this predefined model [here](../../../co

There is a trade-off in this simple model: training with a larger `object_weight` in the loss parameters may result in more false positives (FP), but it will improve accuracy. You can also use `use_nms: True` in the `head_params` to enable NMS which can reduce FP, but it will also reduce TP for close neighbors.

For larger heatmaps and improved accuracy, you can adjust the `attach_index` in the `head_params` to a lower value. This will connect the head to an earlier layer in the backbone, resulting in larger heatmaps. However, be aware that this may lead to slower inference times.

### **Components**

| Name | Alias | Function |
Expand Down
10 changes: 5 additions & 5 deletions luxonis_train/config/predefined_models/detection_fomo_model.py
Original file line number Diff line number Diff line change
Expand Up @@ -25,16 +25,16 @@ class FOMOVariant(BaseModel):
def get_variant(variant: VariantLiteral) -> FOMOVariant:
"""Returns the specific variant configuration for the FOMOModel."""
variants = {
"heavy": FOMOVariant(
backbone="MobileNetV2",
head_params={"num_conv_layers": 2, "conv_channels": 16},
backbone_params={},
),
"light": FOMOVariant(
backbone="EfficientRep",
head_params={"num_conv_layers": 2, "conv_channels": 16},
backbone_params={"variant": "n"},
),
"heavy": FOMOVariant(
backbone="MobileNetV2",
head_params={"num_conv_layers": 2, "conv_channels": 16},
backbone_params={},
),
}

if variant not in variants:
Expand Down

0 comments on commit 2ceacc5

Please sign in to comment.