EfficientDet¶
ConvNet One-Stage Detector
- class lucid.models.EfficientDet(config: EfficientDetConfig)¶
The EfficientDet class implements a family of Efficient Object Detection models, based on the architecture proposed by Tan et al. (2020).
It combines EfficientNet backbones with BiFPN (Bidirectional Feature Pyramid Networks) for scalable and efficient multi-scale detection. Model structure is defined through EfficientDetConfig.
%%{init: {"flowchart":{"curve":"monotoneX","nodeSpacing":50,"rankSpacing":50}} }%%
flowchart LR
linkStyle default stroke-width:2.0px
subgraph sg_m0["<span style='font-size:20px;font-weight:700'>efficientdet_d0</span>"]
style sg_m0 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
m1(["Conv2d x 4<br/><span style='font-size:11px;color:#c53030;font-weight:400'>(1,40,28,28) → (1,64,28,28)</span>"]);
subgraph sg_m2["conv7"]
style sg_m2 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
m3["ReLU"];
m4["Conv2d<br/><span style='font-size:11px;color:#c53030;font-weight:400'>(1,64,4,4) → (1,64,2,2)</span>"];
end
subgraph sg_m5["bifpn"]
style sg_m5 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
subgraph sg_m6["_BiFPN x 2"]
direction TB;
style sg_m6 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
m6_in(["Input"]);
m6_out(["Output"]);
style m6_in fill:#e2e8f0,stroke:#64748b,stroke-width:1px;
style m6_out fill:#e2e8f0,stroke:#64748b,stroke-width:1px;
subgraph sg_m7["convs"]
direction TB;
style sg_m7 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
m8(["_ConvBlock x 8"]);
end
subgraph sg_m9["ups"]
direction TB;
style sg_m9 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
m10(["Upsample x 4<br/><span style='font-size:11px;color:#b83280;font-weight:400'>(1,64,14,14) → (1,64,28,28)</span>"]);
end
subgraph sg_m11["downs"]
direction TB;
style sg_m11 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
m12(["AvgPool2d x 4<br/><span style='font-size:11px;color:#b7791f;font-weight:400'>(1,64,28,28) → (1,64,14,14)</span>"]);
end
m13["ParameterDict"];
subgraph sg_m14["acts"]
direction TB;
style sg_m14 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
m15(["Swish x 8"]);
end
end
end
subgraph sg_m16["_BBoxRegressor"]
style sg_m16 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
subgraph sg_m17["layers"]
direction TB;
style sg_m17 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
subgraph sg_m18["DepthSeparableConv2d"]
direction TB;
style sg_m18 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
m19(["Conv2d x 2"]);
end
m20["BatchNorm2d"];
m21["Swish"];
subgraph sg_m22["DepthSeparableConv2d"]
direction TB;
style sg_m22 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
m23(["Conv2d x 2"]);
end
m24["BatchNorm2d"];
m25["Swish"];
subgraph sg_m26["DepthSeparableConv2d"]
direction TB;
style sg_m26 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
m27(["Conv2d x 2"]);
end
m28["BatchNorm2d"];
m29["Swish"];
end
m30["Conv2d<br/><span style='font-size:11px;color:#c53030;font-weight:400'>(1,64,2,2) → (1,36,2,2)</span>"];
end
subgraph sg_m31["_Classifier"]
style sg_m31 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
subgraph sg_m32["layers"]
direction TB;
style sg_m32 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
subgraph sg_m33["DepthSeparableConv2d"]
direction TB;
style sg_m33 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
m34(["Conv2d x 2"]);
end
m35["BatchNorm2d"];
m36["Swish"];
subgraph sg_m37["DepthSeparableConv2d"]
direction TB;
style sg_m37 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
m38(["Conv2d x 2"]);
end
m39["BatchNorm2d"];
m40["Swish"];
subgraph sg_m41["DepthSeparableConv2d"]
direction TB;
style sg_m41 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
m42(["Conv2d x 2"]);
end
m43["BatchNorm2d"];
m44["Swish"];
end
m45["Conv2d<br/><span style='font-size:11px;color:#c53030;font-weight:400'>(1,64,2,2) → (1,720,2,2)</span>"];
m46["Sigmoid"];
end
m47["_Anchors<br/><span style='font-size:11px;font-weight:400'>(1,3,224,224) → (1,9441,4)</span>"];
m48["_BBoxTransform"];
m49["_ClipBoxes"];
m50["_FocalLoss"];
subgraph sg_m51["_EfficientNetBackbone"]
style sg_m51 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
subgraph sg_m52["model"]
direction TB;
style sg_m52 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
subgraph sg_m53["Sequential"]
direction TB;
style sg_m53 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
m54["Conv2d<br/><span style='font-size:11px;color:#c53030;font-weight:400'>(1,3,224,224) → (1,32,112,112)</span>"];
m55["BatchNorm2d"];
end
subgraph sg_m56["Sequential x 6"]
direction TB;
style sg_m56 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
m56_in(["Input"]);
m56_out(["Output"]);
style m56_in fill:#e2e8f0,stroke:#64748b,stroke-width:1px;
style m56_out fill:#e2e8f0,stroke:#64748b,stroke-width:1px;
m57["_MBConv<br/><span style='font-size:11px;font-weight:400'>(1,32,112,112) → (1,16,112,112)</span>"];
end
end
end
end
input["Input<br/><span style='font-size:11px;color:#a67c00;font-weight:400'>(1,3,224,224)</span>"];
output["Output<br/><span style='font-size:11px;color:#a67c00;font-weight:400'>(1,9441,80)x3</span>"];
style input fill:#fff3cd,stroke:#a67c00,stroke-width:1px;
style output fill:#fff3cd,stroke:#a67c00,stroke-width:1px;
style m1 fill:#ffe8e8,stroke:#c53030,stroke-width:1px;
style m3 fill:#faf5ff,stroke:#6b46c1,stroke-width:1px;
style m4 fill:#ffe8e8,stroke:#c53030,stroke-width:1px;
style m10 fill:#fdf2f8,stroke:#b83280,stroke-width:1px;
style m12 fill:#fefcbf,stroke:#b7791f,stroke-width:1px;
style m15 fill:#faf5ff,stroke:#6b46c1,stroke-width:1px;
style m19 fill:#ffe8e8,stroke:#c53030,stroke-width:1px;
style m20 fill:#e6fffa,stroke:#2c7a7b,stroke-width:1px;
style m21 fill:#faf5ff,stroke:#6b46c1,stroke-width:1px;
style m23 fill:#ffe8e8,stroke:#c53030,stroke-width:1px;
style m24 fill:#e6fffa,stroke:#2c7a7b,stroke-width:1px;
style m25 fill:#faf5ff,stroke:#6b46c1,stroke-width:1px;
style m27 fill:#ffe8e8,stroke:#c53030,stroke-width:1px;
style m28 fill:#e6fffa,stroke:#2c7a7b,stroke-width:1px;
style m29 fill:#faf5ff,stroke:#6b46c1,stroke-width:1px;
style m30 fill:#ffe8e8,stroke:#c53030,stroke-width:1px;
style m34 fill:#ffe8e8,stroke:#c53030,stroke-width:1px;
style m35 fill:#e6fffa,stroke:#2c7a7b,stroke-width:1px;
style m36 fill:#faf5ff,stroke:#6b46c1,stroke-width:1px;
style m38 fill:#ffe8e8,stroke:#c53030,stroke-width:1px;
style m39 fill:#e6fffa,stroke:#2c7a7b,stroke-width:1px;
style m40 fill:#faf5ff,stroke:#6b46c1,stroke-width:1px;
style m42 fill:#ffe8e8,stroke:#c53030,stroke-width:1px;
style m43 fill:#e6fffa,stroke:#2c7a7b,stroke-width:1px;
style m44 fill:#faf5ff,stroke:#6b46c1,stroke-width:1px;
style m45 fill:#ffe8e8,stroke:#c53030,stroke-width:1px;
style m46 fill:#faf5ff,stroke:#6b46c1,stroke-width:1px;
style m54 fill:#ffe8e8,stroke:#c53030,stroke-width:1px;
style m55 fill:#e6fffa,stroke:#2c7a7b,stroke-width:1px;
input --> m54;
m1 --> m3;
m10 -.-> m8;
m12 -.-> m8;
m15 --> m10;
m15 --> m12;
m15 --> m6_out;
m19 --> m20;
m20 --> m21;
m21 --> m23;
m23 --> m24;
m24 --> m25;
m25 --> m27;
m27 --> m28;
m28 --> m29;
m29 --> m30;
m3 --> m4;
m30 -.-> m19;
m30 --> m47;
m34 --> m35;
m35 --> m36;
m36 --> m38;
m38 --> m39;
m39 --> m40;
m4 -.-> m15;
m40 --> m42;
m42 --> m43;
m43 --> m44;
m44 --> m45;
m45 --> m46;
m46 -.-> m19;
m46 -.-> m34;
m47 --> output;
m54 --> m55;
m55 -.-> m57;
m56_in -.-> m57;
m56_out --> m1;
m56_out -.-> m56_in;
m57 -.-> m56_in;
m57 --> m56_out;
m6_in -.-> m8;
m6_out -.-> m34;
m8 -.-> m15;
m8 --> m6_in;
Class Signature¶
class EfficientDet(nn.Module):
def __init__(self, config: EfficientDetConfig) -> None
Parameters¶
config (EfficientDetConfig): Configuration object describing the compound scaling coefficient, anchor count, and class count used to build the EfficientDet variant.
Backbone Network¶
The backbone network of EfficientDet, a truncated EfficientNet model can be accessed via .backbone attribute of an instance of the class EfficientDet.
Warning
EfficientNet backbone for EfficientDet model is not pre-trained by default.
The user should pre-train the corresponding separate EfficientNet variant model for image classification task and then migrate the weights of stage 1-7 to EfficientDet.backbone.model.
Weight migration can be easily done with state_dict and load_state_dict methods.
pretrained = efficientnet_b0(...) # Pre-trained for Image Classification
model = EfficientDet(EfficientDetConfig(...))
# Migrate Weights Stage-by-Stage
for i in range(7):
# Save state-dict of individual pre-trained model stage
st_dict = getattr(pretrained, f"stage{i + 1}").state_dict()
# model.backbone.model is an nn.Sequential module with 7 stages
model.backbone.model[i].load_state_dict(st_dict)
Input Format¶
The model expects a 4D input tensor of shape:
(N, 3, H, W)
Where:
N is the batch size,
3 represents the RGB image channels,
H, W are the input height and width (variable depending on compound_coef).
Target Format¶
Targets should be provided as a list of Tensors, one per image:
targets = [Tensor_i of shape (Ni, 5)]
Where:
Ni is the number of objects in the i-th image,
each row of shape (5,) corresponds to [x_min, y_min, x_max, y_max, class_id].
Note
Bounding box coordinates are expected in absolute pixel units, not normalized.
Loss Computation¶
During training, the total loss combines classification and box regression terms:
Where:
\(\mathcal{L}_{cls}\) is the Focal Loss for classification,
\(\mathcal{L}_{box}\) is the Smooth L1 or IoU-based regression loss.
Focal Loss Explanation¶
The Focal Loss addresses the issue of class imbalance by down-weighting easy examples and focusing training on hard, misclassified samples.
It is defined as:
Where:
\(p_t\) is the predicted probability for the true class:
\[\begin{split}p_t = \begin{cases} p, & \text{if } y = 1 \\ 1 - p, & \text{otherwise} \end{cases}\end{split}\]\(\alpha_t\) is a weighting factor balancing positive and negative samples,
\(\gamma\) (the focusing parameter) controls how much the easy examples are down-weighted.
The gradient of the Focal Loss with respect to the input probability \(p\) is:
As \(p_t \to 1\), the gradient approaches zero, reducing the contribution of well-classified examples and allowing the model to focus on harder ones.
Tip
Typical values: \(\alpha_t = 0.25\), \(\gamma = 2.0\).
For detection, Focal Loss is applied to every anchor across all pyramid levels.
Note
The Focal Loss was first introduced in Lin et al., 2017 (RetinaNet), and EfficientDet adopted it to stabilize classification across feature scales.
Methods¶
Tip
EfficientDet uses compound scaling to balance model accuracy and efficiency: larger compound_coef values correspond to deeper and wider networks (D0-D7).
Warning
The target list length must match the batch size, and each Tensor inside should contain [x_min, y_min, x_max, y_max, class_id] for all objects in that image.