HDLBits Circuits 状态机专栏(二)

Circuits -->Sequential Logic -->Finite State Machines

module top_module (
	input clk,
	input in,
	input areset,
	output out
);

	// Give state names and assignments. I'm lazy, so I like to use decimal numbers.
	// It doesn't really matter what assignment is used, as long as they're unique.
	parameter A=0, B=1;
	reg state;		// Ensure state and next are big enough to hold the state encoding.
	reg next;
    
    
    // A finite state machine is usually coded in three parts:
    //   State transition logic
    //   State flip-flops
    //   Output logic
    // It is sometimes possible to combine one or more of these blobs of code
    // together, but be careful: Some blobs are combinational circuits, while some
    // are clocked (DFFs).
    
    
    // Combinational always block for state transition logic. Given the current state and inputs,
    // what should be next state be?
    // Combinational always block: Use blocking assignments.
    always@(*) begin
		case (state)
			A: next = in ? A : B;
			B: next = in ? B : A;
		endcase
    end
    
    
    
    // Edge-triggered always block (DFFs) for state flip-flops. Asynchronous reset.
    always @(posedge clk, posedge areset) begin
		if (areset) state <= B;		// Reset to state B
        else state <= next;			// Otherwise, cause the state to transition
	end
		
		
		
	// Combinational output logic. In this problem, an assign statement is the simplest.
	// In more complex circuits, a combinational always block may be more suitable.
	assign out = (state==B);

	
endmodule

(这里不是自己写的,自己写的没保存找不到了)

// Note the Verilog-1995 module declaration syntax here:
module top_module(clk, reset, in, out);
    input clk;
    input reset;    // Synchronous reset to state B
    input in;
    output out;//  
    reg out;

    // Fill in state name declarations
	localparam A=1'b0,B=1'b1;
    reg present_state, next_state;

    always @(posedge clk) begin
        if (reset) begin  
            // Fill in reset logic
            present_state <= B;
        end
        else 
            // State flip-flops
             present_state = next_state;   
    end
    
    always @(*) begin
            case (present_state)
                // Fill in state transition logic
                A:	next_state <= in?A:B;
                B:	next_state <= in?B:A;
            endcase
    end

    always @(posedge clk or posedge reset) begin
            case (present_state)
                // Fill in output logic
                A: out=1'b0;
                B: out=1'b1;
            endcase
    end

endmodule

module top_module(
    input clk,
    input areset,    // Asynchronous reset to OFF
    input j,
    input k,
    output out); //  

    parameter OFF=0, ON=1; 
    reg state, next_state;

    always@(posedge clk, posedge areset)begin
        if(areset)
            state<=OFF;
        else
            state<=next_state;
    end
    
    always @(*) begin
        // State transition logic
        case(state)
            ON:	next_state=k?OFF:ON;
            OFF: next_state=j?ON:OFF;
            default:next_state=0;
        endcase
    end

    always @(posedge clk, posedge areset) begin
        // State flip-flops with asynchronous reset
        if(areset)
            out<=OFF;
        else
            case(next_state)
                ON:	out<=1;
                OFF: out<=0;
                default:next_state=0;
            endcase
    end

    // Output logic
    // assign out = (state == ...);

endmodule

这里always里面的电平敏感  always@(posedge clk, posedge areset)和上一个always@(posedge clk)结果不一样,应该是同步复位和异步复位的区别导致的

module top_module(
    input clk,
    input reset,    // Synchronous reset to OFF
    input j,
    input k,
    output out); //  

    parameter OFF=0, ON=1; 
    reg state, next_state;

    always@(posedge clk)begin
        if(reset)
            state<=OFF;
        else
            state<=next_state;
    end
    
    always @(*) begin
        // State transition logic
        case(state)
            ON:	next_state=k?OFF:ON;
            OFF: next_state=j?ON:OFF;
            default:next_state=0;
        endcase
    end

    always @(posedge clk) begin
        // State flip-flops with asynchronous reset
        if(reset)
            out<=OFF;
        else
            case(next_state)
                ON:	out<=1;
                OFF: out<=0;
                default:next_state=0;
            endcase
    end

    // Output logic
    // assign out = (state == ...);

endmodule

module top_module(
    input in,
    input [1:0] state,
    output [1:0] next_state,
    output out); //

    parameter A=2'b00, B=2'b01, C=2'b10, D=2'b11;
    
    always @(*)begin
        case(state)
            A:begin
                if(in)
                    next_state=B;
                else
                    next_state=A;
            end
            B:begin
                if(in)
                    next_state=B;
                else
                    next_state=C;
            end
            C:begin
                if(in)
                    next_state=D;
                else
                    next_state=A;
            end
            D:begin
                if(in)
                    next_state=B;
                else
                    next_state=C;
            end
            default:next_state=A;
        endcase
    end

    // State transition logic: next_state = f(state, in
### RT-DETRv3 网络结构分析 RT-DETRv3 是一种基于 Transformer 的实时端到端目标检测算法,其核心在于通过引入分层密集正监督方法以及一系列创新性的训练策略,解决了传统 DETR 模型收敛慢和解码器训练不足的问题。以下是 RT-DETRv3 的主要网络结构特点: #### 1. **基于 CNN 的辅助分支** 为了增强编码器的特征表示能力,RT-DETRv3 引入了一个基于卷积神经网络 (CNN) 的辅助分支[^3]。这一分支提供了密集的监督信号,能够与原始解码器协同工作,从而提升整体性能。 ```python class AuxiliaryBranch(nn.Module): def __init__(self, in_channels, out_channels): super(AuxiliaryBranch, self).__init__() self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1) self.bn = nn.BatchNorm2d(out_channels) def forward(self, x): return F.relu(self.bn(self.conv(x))) ``` 此部分的设计灵感来源于传统的 CNN 架构,例如 YOLO 系列中的 CSPNet 和 PAN 结构[^2],这些技术被用来优化特征提取效率并减少计算开销。 --- #### 2. **自注意力扰动学习策略** 为解决解码器训练不足的问题,RT-DETRv3 提出了一种名为 *self-att 扰动* 的新学习策略。这种策略通过对多个查询组中阳性样本的标签分配进行多样化处理,有效增加了阳例的数量,进而提高了模型的学习能力和泛化性能。 具体实现方式是在训练过程中动态调整注意力权重分布,确保更多的高质量查询可以与真实标注 (Ground Truth) 进行匹配。 --- #### 3. **共享权重解编码器分支** 除了上述改进外,RT-DETRv3 还引入了一个共享权重的解编码器分支,专门用于提供密集的正向监督信号。这一设计不仅简化了模型架构,还显著降低了参数量和推理时间,使其更适合实时应用需求。 ```python class SharedDecoderEncoder(nn.Module): def __init__(self, d_model, nhead, num_layers): super(SharedDecoderEncoder, self).__init__() decoder_layer = nn.TransformerDecoderLayer(d_model=d_model, nhead=nhead) self.decoder = nn.TransformerDecoder(decoder_layer, num_layers=num_layers) def forward(self, tgt, memory): return self.decoder(tgt=tgt, memory=memory) ``` 通过这种方式,RT-DETRv3 实现了高效的目标检测流程,在保持高精度的同时大幅缩短了推理延迟。 --- #### 4. **与其他模型的关系** 值得一提的是,RT-DETRv3 并未完全抛弃经典的 CNN 技术,而是将其与 Transformer 结合起来形成混合架构[^4]。例如,它采用了 YOLO 系列中的 RepNCSP 模块替代冗余的多尺度自注意力层,从而减少了不必要的计算负担。 此外,RT-DETRv3 还借鉴了 DETR 的一对一匹配策略,并在此基础上进行了优化,进一步提升了小目标检测的能力。 --- ### 总结 综上所述,RT-DETRv3 的网络结构主要包括以下几个关键组件:基于 CNN 的辅助分支、自注意力扰动学习策略、共享权重解编码器分支以及混合编码器设计。这些技术创新共同推动了实时目标检测领域的发展,使其在复杂场景下的表现更加出色。 ---
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值