---
title: Inverted Residual Depthwise Convolution Block
url: https://www.emergentmind.com/topics/inverted-residual-depthwise-convolution-block-irdcb
type: topic
---

# Inverted Residual Depthwise Convolution Block

The **Inverted Residual Depthwise Convolution Block (IRDCB)** is a lightweight convolutional building unit that combines an **inverted residual structure**, **depthwise-separable computation**, and a **conditional residual connection**. In the specific usage documented for **HierLight-YOLO**, IRDCB is introduced as a replacement for the original YOLOv8-style **C2f module** in order to reduce parameters and computational cost while preserving detection accuracy in UAV imagery dominated by small objects [2509.22365]. More broadly, IRDCB belongs to the same architectural family as the **MobileNetV2 inverted residual block**, whose defining pattern is expansion in a higher-dimensional latent space, depthwise spatial filtering, and projection back to a thin bottleneck representation [1801.04381]. Related works also explore alternative orderings and reallocations of the same basic ingredients, including the **Depthwise-Pointwise-Depthwise (DPD) inverted bottleneck block** and the **asymmetrical bottleneck** [1909.01026], [2104.07770].

## 1. Terminological scope and architectural lineage

The term **IRDCB** is used explicitly in **HierLight-YOLO**, where it denotes one of the two “innovative lightweight modules” used to improve efficiency in a YOLOv8-based detector for UAV photography [2509.22365]. In that paper, IRDCB is presented as a redesign of the backbone feature-extraction block, intended to reduce model size and FLOPs without sacrificing the fine-grained spatial information needed for small-object detection.

At the level of architectural lineage, IRDCB is directly connected to **MobileNetV2**, whose central building unit is described as an **inverted residual block with linear bottleneck** [1801.04381]. MobileNetV2 inverts the classical bottleneck pattern: instead of connecting wide tensors and compressing internally, it starts from a thin bottleneck tensor, expands it, applies the nonlinear transform in the expanded space, then projects back to a thin output, with shortcut connections between the thin bottlenecks.

The broader family of related blocks includes variants that preserve the inverted-bottleneck idea while modifying internal operator order or the allocation of computation. The **DPD block** replaces pointwise expansion with **depthwise convolution with channel multiplier \(m>1\)**, yielding the sequence **DWC \((m>1)\) \(\to\) PWC \(\to\) DWC \((m=1)\)** [1909.01026]. **AsymmNet** keeps the standard pointwise–depthwise–pointwise macro-structure but changes the balance between the two pointwise layers through feature reuse and an asymmetry rate \(r\) [2104.07770].

This suggests that “IRDCB” is best understood as a specific named instance within a larger class of **inverted residual / depthwise-convolution bottleneck blocks**, rather than as a universally standardized term. In the available sources, the exact name is tied to HierLight-YOLO, whereas MobileNetV2 supplies the canonical template and later works supply closely related variants.

## 2. Canonical block structure

In **MobileNetV2**, the practical inverted residual block has the tensor transformation

\[
h \times w \times k \xrightarrow{1\times 1} h \times w \times (tk) \xrightarrow{3\times 3,\ s=s} \frac{h}{s} \times \frac{w}{s} \times (tk) \xrightarrow{\text{linear }1\times 1} \frac{h}{s} \times \frac{w}{s} \times k'
\]

where the first \(1\times1\) convolution expands channels by factor \(t\), the middle \(3\times3\) operator is depthwise, and the final \(1\times1\) projection is linear [1801.04381]. The default expansion factor reported for MobileNetV2 is **\(t=6\)**.

In **HierLight-YOLO**, the IRDCB is also a three-stage transform, but its middle stage is deeper. The block takes an input tensor \(\mathbf{x}_{\text{in}}\) with input channels \(c_1\) and output channels \(c_2\), and is described as

\[
\mathcal{F}(\mathbf{x}_{\text{in}}) =
\text{Conv}_{1\times1}^{\downarrow}_{\text{Compress}}
\circ \text{DWConv}_{3\times3}
\circ \text{DWConv}_{3\times3}
\circ \text{Conv}_{1\times1}^{\uparrow}_{\text{Expand}}
(\mathbf{x}_{\text{in}})
\]

with the intended operational sequence stated as:

1. **Expand** with a \(1\times1\) convolution  
2. **Filter** with two stacked \(3\times3\) depthwise convolutions  
3. **Compress** with a final \(1\times1\) convolution  
4. **Add a residual shortcut** if input and output channels match [2509.22365]

The expansion stage increases channels from \(c_1\) to

\[
c^* = \lfloor c_1 t \rfloor,\quad t \ge 1
\]

The two depthwise convolutions each use a **\(3\times3\)** kernel, **stride 1**, **padding 1**, and **group number \(= c^*\)**, so each channel is filtered independently. The compression stage maps \(c^*\) back to \(c_2\) using another \(1\times1\) convolution [2509.22365].

Relative to MobileNetV2, the distinguishing change is therefore not the presence of the inverted residual principle itself, but the use of **two consecutive depthwise convolutions instead of one**. The paper states that this improves nonlinear representation and local pattern capture [2509.22365].

## 3. Computational rationale and efficiency model

The efficiency logic underlying IRDCB-style blocks comes from the separation of **channel mixing** and **spatial filtering**. In a standard convolution block, a regular \(k\times k\) convolution performs both simultaneously, so each output channel interacts with all input channels. By contrast, inverted residual depthwise blocks use **\(1\times1\) pointwise convolutions** for channel expansion and compression, while using **depthwise \(3\times3\) convolutions** for spatial filtering [1801.04381], [2509.22365].

For **MobileNetV2**, the paper gives the cost of depthwise separable convolution as

\[
h_i \cdot w_i \cdot d_i (k^2 + d_j)
\]

compared with standard convolution cost

\[
h_i \cdot w_i \cdot d_i \cdot d_j \cdot k \cdot k
\]

and states that for \(k=3\), depthwise separable convolutions are roughly **8–9× cheaper** than full convolutions, with only a small accuracy drop [1801.04381].

For a MobileNetV2 bottleneck block with input size \(h \times w\), expansion factor \(t\), kernel size \(k\), input channels \(d'\), and output channels \(d''\), the multiply-add cost is

\[
h \cdot w \cdot d' \cdot t(d' + k^2 + d'')
\]

which formalizes why the expensive channel-mixing operations are confined to \(1\times1\) layers while the spatial operator remains depthwise [1801.04381].

In **HierLight-YOLO**, the IRDCB efficiency argument is stated more directly through complexity comparisons. The expansion \(1\times1\) convolution has cost \(c_1 c^* HW\), whereas a standard convolution at the same channel dimensions would cost \(k^2 c_1 c^* HW\). Each depthwise convolution has complexity

\[
c^* k^2 HW
\]

whereas a standard convolution over the expanded channels would cost

\[
c^{*2} k^2 HW
\]

This is identified as the core source of savings [2509.22365].

A closely related efficiency argument appears in the **DPD block** literature. For a feature map \(F\in \mathbb{R}^{W\times H\times C}\) expanded to \(G\in \mathbb{R}^{W\times H\times mC}\), a **PWC** needs \(C\cdot mC\) parameters and computational cost \(W\cdot H\cdot C\cdot mC\), whereas a **DWC** with kernel size \(k\) and output channels \(mC\) needs \(k^2\cdot mC\) parameters and cost \(W\cdot H\cdot k^2\cdot mC\). The parameter and computation ratios are both

\[
\frac{C}{k^2}
\]

and for \(k=3\), the argument is that depthwise convolution is much more efficient than pointwise convolution for channel expansion when \(C\) is much larger than \(9\) [1909.01026].

Taken together, these formulations support a consistent interpretation: the practical advantage of IRDCB-style blocks comes from delegating spatial modeling to depthwise operators and restricting dense cross-channel mixing to \(1\times1\) layers, thereby reducing parameters, FLOPs, and activation cost.

## 4. Residual topology, bottlenecks, and information flow

A defining property of inverted residual blocks is that the shortcut operates between **thin bottleneck tensors**, not between wide expanded tensors. **MobileNetV2** states that “the shortcut connections are between the thin bottleneck layers” and that shortcuts connecting bottlenecks perform better than shortcuts connecting expanded layers [1801.04381]. Residual addition is used only when the block preserves both spatial resolution and channel count; in practice, if stride \(s=1\) and input/output channels match, then

\[
y = x + F(x)
\]

otherwise there is no residual addition [1801.04381].

The **IRDCB** in HierLight-YOLO follows the same conditional logic, formulated explicitly as

\[
\mathbf{x}_{\text{out}} =
\begin{cases}
\mathbf{x}_{\text{in}} + \mathcal{F}(\mathbf{x}_{\text{in}}), & \text{if } c_1 = c_2 \\
\mathcal{F}(\mathbf{x}_{\text{in}}), & \text{otherwise}
\end{cases}
\]

Thus, the block behaves as a residual unit only when channel dimensions are compatible [2509.22365].

MobileNetV2 also links this residual topology to the concept of **linear bottlenecks**. The paper argues that applying nonlinearity in a narrow space can collapse information lying on a low-dimensional manifold, and therefore “it is important to remove non-linearities in the narrow layers in order to maintain representational power” [1801.04381]. Its core design principle is therefore **nonlinearity in the wide layer** and **no nonlinearity in the narrow layer**.

No explicit activation function or normalization layer is specified in the IRDCB subsection of HierLight-YOLO. The paper notes that such layers may exist implicitly in a YOLOv8 framework, but they are not explicitly defined in the IRDCB description [2509.22365]. Accordingly, the available evidence supports a structural characterization of IRDCB, but not a more detailed claim about its activation or normalization policy.

A common misconception is to treat all inverted residual blocks as interchangeable. The sources do not support that view. MobileNetV2’s canonical block, HierLight-YOLO’s IRDCB, the DPD block, and AsymmNet’s asymmetrical bottleneck all share the inverted-residual/depthwise design space, but they differ materially in the placement of depthwise operators, the allocation of channel expansion, and the precise role of residual and projection layers [1801.04381], [1909.01026], [2104.07770], [2509.22365].

## 5. Major variants and closely related blocks

Several architectures modify the inverted residual depthwise block while preserving its lightweight design objective.

| Block | Core operator order | Distinctive feature |
|---|---|---|
| MobileNetV2 inverted residual | \(1\times1\) expand \(\to\) \(3\times3\) DW \(\to\) linear \(1\times1\) projection | Linear bottleneck and shortcut between thin layers |
| DPD block | DWC \((m>1)\) \(\to\) PWC \(\to\) DWC \((m=1)\) | Uses depthwise convolution for channel expansion |
| Asymmetrical bottleneck | PW1 adjustment \(\to\) feature reuse/Concat \(\to\) DW \(\to\) PW2 | Reallocates computation from PW1 to PW2 |
| IRDCB in HierLight-YOLO | \(1\times1\) expand \(\to\) DW \(\to\) DW \(\to\) \(1\times1\) compress | Uses two consecutive depthwise convolutions |

The **DPD block** is the closest case in which depthwise convolution is promoted from a middle spatial operator to the expansion mechanism itself. The first DWC uses channel multiplier \(m>1\) to increase channels and complete downsampling, the PWC combines information of different channels and decreases the number of channels, and the final DWC filters each channel [1909.01026]. The paper concludes that “extracting spatial features is more important than combining channel information” and reports that networks with more DWC layers outperform networks with more \(1\times1\) convolution layers.

**AsymmNet** instead revisits the two pointwise convolutions in a standard MobileNetV2-style block. It proposes to reduce the first pointwise convolution, enrich information flow by direct feature reuse, and migrate the saved computations to the second pointwise convolution [2104.07770]. The resulting block keeps the familiar inverted-residual macro-structure but changes the internal balance of capacity and channel mixing. For stride \(s=1\), the theoretical cost ratio is given as

\[
R_c = \frac{hwc(tc-rc)+k^2hw(tc+rc)+hw(tc+rc)c}
{hwc(tc)+k^2hw(tc)+hw(tc)c}
\]

which simplifies to

\[
R_c = 1+\frac{rk^2}{2tc+k^2t} \approx 1
\]

The stated implication is that computation can be reallocated while total cost remains approximately unchanged [2104.07770].

These variants clarify that the design space around IRDCB is not limited to a single operator sequence. Different works target different bottlenecks: MobileNetV2 emphasizes information preservation in narrow layers, DPD emphasizes replacing pointwise expansion with depthwise expansion, AsymmNet emphasizes asymmetrical allocation between the two pointwise layers, and HierLight-YOLO emphasizes stronger local pattern capture through two stacked depthwise convolutions.

## 6. Empirical performance and documented trade-offs

The most direct empirical evidence for **IRDCB** comes from **HierLight-YOLO**. In the module comparison reported in Table 4, the results are:

| Module | Accuracy / AP | Model size / cost |
|---|---|---|
| C3 | 44.1 AP\(_{0.5}\) | 9.8M params, 36.0 GFLOPs |
| C2f | 44.9 AP\(_{0.5}\) | 11.3M params, 38.1 GFLOPs |
| IRDCB | 44.8 AP\(_{0.5}\) | 8.8M params, 34.5 GFLOPs |

The paper summarizes this as **22.1% fewer parameters than the original C2f module**, **9.4% fewer FLOPs than C2f**, and **no meaningful accuracy loss** [2509.22365]. Relative to C2f, the AP\(_{0.5}\) difference is only **0.1%**, while parameters fall from **11.3M** to **8.8M** and FLOPs from **38.1G** to **34.5G**.

The full ablation chain reports that baseline YOLOv8s achieves **43.0 AP\(_{0.5}\)**, **26.0 AP\(_{0.5:0.95}\)**, **11.1M params**, and **28.5 GFLOPs**, while the version with **P2 head + HEPAN + IRDCB** reaches **47.1 AP\(_{0.5}\)**, **29.0 AP\(_{0.5:0.95}\)**, **8.8M params**, and **34.5 GFLOPs** [2509.22365]. The paper specifically notes that IRDCB reduces model size while maintaining comparable accuracy.

The IRDCB expansion study identifies **\(n=2, t=2\)** as the best setting, with **44.9 AP\(_{0.5}\)**, **8.8M params**, and **34.5 GFLOPs**. Increasing \(t\) from **2** to **4** with fixed \(n=1\) yields only a small gain from **44.0** to **44.6 AP\(_{0.5}\)**, while parameters rise from **8.8M** to **9.4M**; for \(n=2\), increasing \(t\) beyond **2** degrades performance from **44.9** at \(t=2\) to **44.5** at \(t=3\) and **44.4** at \(t=4\) [2509.22365]. The paper concludes that too much expansion hurts optimization and efficiency.

The broader literature supports the general utility of this architectural family. **MobileNetV2** reports a main model with **3.4M parameters** and **300M multiply-adds**, and states that it achieves better accuracy than MobileNetV1 at comparable or lower cost [1801.04381]. **DPDNet** reports that the number of parameters is only about **60%** of that of MobileNetV2 for networks with the same number of layers, while achieving approximated accuracy [1909.01026]. In one representative setting, DPDNet uses **0.09M parameters and 12.6M FLOPs** versus **0.14M parameters and 23.3M FLOPs** for MobileNetV2, with **90.42%** on CIFAR-10 and **66.36%** on CIFAR-100 compared with MobileNetV2’s **89.92%** and **67.07%** [1909.01026].

These results do not imply that all IRDCB-like blocks dominate all alternatives on all tasks. Rather, they document a recurring trade-off: lightweight inverted residual depthwise blocks can reduce parameters and FLOPs substantially while maintaining similar accuracy, but the optimal internal expansion and operator ordering remain architecture- and task-dependent.

## 7. Interpretation, significance, and recurrent points of debate

The principal significance of IRDCB lies in its role as a **lightweight feature-extraction block** that seeks to preserve representational adequacy under tight compute and memory budgets. In the detection setting of HierLight-YOLO, the block is motivated by the need to remain efficient on resource-constrained platforms without losing the spatial detail required for UAV small-object detection [2509.22365]. In the mobile-classification setting of MobileNetV2, the same family is motivated by the need to separate input/output bottleneck dimension from the expressiveness of the nonlinear transformation [1801.04381].

One recurrent point of debate concerns the relative importance of **spatial extraction** versus **channel mixing**. The DPD paper explicitly argues that “extracting spatial features is more important than combining channel information” and uses the performance of depthwise-heavy variants to support that interpretation [1909.01026]. By contrast, AsymmNet argues that the second pointwise convolution is especially important for learning cross-channel feature correlations, and therefore computation should be migrated toward PW2 rather than left in PW1 [2104.07770]. These positions are not identical; they reflect different emphases within the same low-cost design space.

Another frequent misunderstanding is to equate “inverted residual” with a single immutable recipe. The available evidence shows at least four distinct realizations: the canonical **MobileNetV2** block with one depthwise layer and linear projection; the **IRDCB** of HierLight-YOLO with two depthwise layers; the **DPD** block with depthwise expansion; and the **AsymmNet** block with asymmetrical pointwise allocation [1801.04381], [2509.22365], [1909.01026], [2104.07770]. Their commonality lies in the use of thin bottlenecks, lightweight spatial filtering, and conditional shortcutting, but their internal mechanics differ enough to affect both efficiency and accuracy.

A plausible implication is that IRDCB is best treated not as a singular fixed artifact, but as a design point within a family of **inverted residual depthwise bottlenecks**. In the sources considered here, its most concrete and fully named instantiation is the YOLOv8-backbone module in HierLight-YOLO, while its conceptual foundations are anchored in MobileNetV2 and its subsequent variants.

Source: https://www.emergentmind.com/topics/inverted-residual-depthwise-convolution-block-irdcb